Jump to ContentJump to Main Navigation
Objects of Metaphor$

Samuel Guttenplan

Print publication date: 2005

Print ISBN-13: 9780199280896

Published to Oxford Scholarship Online: July 2005

DOI: 10.1093/0199280894.001.0001

Show Summary Details
Page of

PRINTED FROM OXFORD SCHOLARSHIP ONLINE (www.oxfordscholarship.com). (c) Copyright Oxford University Press, 2019. All Rights Reserved. An individual user may print out a PDF of a single chapter of a monograph in OSO for personal use.  Subscriber: null; date: 17 September 2019

The Semantic Descent Account

The Semantic Descent Account

Chapter:
(p.93) 3 The Semantic Descent Account
Source:
Objects of Metaphor
Author(s):

Samuel Guttenplan (Contributor Webpage)

Publisher:
Oxford University Press
DOI:10.1093/0199280894.003.0004

Abstract and Keywords

The notion of semantic descent made familiar by Quine is extended to a movement from the first-floor level of language use to the level of objects that language typically describes; descent here is to a basement level. The idea of such a descent is combined with the idea of qualification to produce what is called the ‘Semantic Descent’ account of metaphor. According to this account, metaphor first requires semantic descent to a level of (appropriate) non-linguistic objects, and these objects then fulfill the predicative function described as qualification. The account is presented in a relatively minimalist way, to show how it copes with the features of metaphor discussed in Chapter 1, as well as provide a clear view of the obvious objections that might be raised against it. The latter centrally includes the fear that the use of objects as qualifiers might be too indeterminate. This is countered with a discussion of a notion called ‘attunement’.

Keywords:   W.V.O. Quine, qualification, attunement, indeterminacy, semantic descent

In the previous chapter, I set out to see whether some of things we happen to do with words could also be done without them, but right at the beginning I cautioned against thinking that the results of that investigation could be immediately applied to metaphor. It is certainly true that framing metaphors is something we do with words, but I didn't actually get further than a consideration of the more basic linguistic functions of reference and predication. Nor is metaphor going to be added to that list here. For, though the idea of a wordless metaphor is perfectly coherent—indeed it is something that will crop up in Chapter 4—the single item on the agenda of this chapter is an account of metaphor as a linguistic phenomenon. Having established that qualification is predication, or is in effect predication, in which objects and not words do the work, we need now to combine that result with the fact that in metaphor things begin with words.

Yet another cautionary note should be inserted here. In what follows, I will set out my account in a fairly minimalist way, saving many elaborations and embellishments for the two chapters which follow. This minimalism will be most evident in the simplicity and paucity of examples. As will I hope become apparent, I have not streamlined my exposition in order illicitly to gain plausibility for the account, though this certainly seems a common enough strategy in the literature.1 But in writing this chapter, I have been keen to keep the account itself, and certain aspects of its defence, in sharp focus. The wide range of features typical of metaphor and the variety and richness of examples can easily overwhelm any exposition. So, to repeat, in this chapter I shall keep things simple, defending the account where necessary, but making sure that its overall structure stands out.

That said, as noted in the Introduction, minimalism has its risks: starkly simple examples can make certain kinds of objection seem pressing, even though, against the background of richer examples and further considerations, many of these objections should fall away. Still, as honesty requires it, I shall flag up these objections in this chapter, while nonetheless inviting you to reserve final judgement until you read Chapters 4 and 5.

(p.94) 3.1. Metaphor and Semantic Descent

The notion of ‘semantic ascent’ is too familiar to need much of an introduction.2 Asserting that the sky is blue by saying ‘The sky is blue’ is using language in its most ordinary way. Call this the ‘ground floor’ use of language. Saying of the sentence ‘The sky is blue’ that it is true achieves largely the same purpose as the original assertion, but it does so by engaging in a bit of semantic ascent; it is moving up a level of language by speaking, not immediately about the sky itself, but about the truth of words which themselves speak of it. Assertions about the truth of sentences are typically first-floor uses of language, being one level up from the ground (at least in Europe). What I claim goes on in metaphorical utterances is a bit of semantic descent, but, perhaps surprisingly, it is descent that begins from the ground floor of language use, and moves down to what we can think of as a sort of ‘basement’. Since I don't expect any of this to be familiar, I shall discuss a couple of examples, in the course of which it should become clear not only what semantic descent is, but how, in allowing us to tap into qualification, it gives us a fresh way of looking at metaphor.

Consider again Romeo's assertion:

(R) Juliet is the sun.

Assuming as I have throughout that (R) has already been identified as metaphorical, the outlines of my proposal are quite simple. When (R) is understood as an ordinary assertion at the ground-floor level of language, it is either false, or perhaps even a bit of nonsense. But I suggest that, instead of pausing over any such reading, we take the metaphoricality of (R) to demand a kind of semantic descent. Instead of thinking of the word ‘sun’ in (R) as a word that plays only its usual natural language role in the predicate ‘is the sun’, think of the object that this word stands in for—think of the sun itself. Both the sun and the word ‘sun’ are objects, albeit of radically different kinds. The one is the fiery nuclear star at the centre of the solar system which supports life on earth; the other a set of marks that play a special role in a complex linguistic practice. Yet focus just on the fact that they are both objects. We do not have to think of the predicate in (R) as simply fulfilling the ground-floor predicate role its word-objects have been assigned in a complex social practice. Think in addition that the function of the word-objects in (R) includes taking us from this ground-floor level to a level below—to the basement-level of non-word-objects. In the specific case of (R), ‘the sun’ is pressed into service within a natural language predicative structure to take us to the sun, and it is the latter object that gives us information about Juliet.3 The hearer is invited to understand this object as a qualifier (p.95) of Juliet in just the way that, in the examples of Chapter 2, swatches, colour cards, dustbins, buildings, fallen trees, and palings qualified their subjects. Hence, though the initial setting is wholly one of natural language, much of the work of (R) is accomplished by means of the not naturally linguistic, though predicative, mechanism of qualification. Since ordinary natural language predication is at the ground-floor level, the move to the level at which qualification figures is a move to a sort of basement level.

Without stopping to discuss this at length here, there is something appealing about the idea that metaphor works in the basement. After all basements are where one finds foundations, and there are those who urge us to see metaphor as somehow implicated in the very foundation of language. Having already suggested that qualification plays such a role in categorization, it is not difficult to see how semantic descent followed by qualification might satisfy that urge.

What I have given so far is only an outline of the semantic descent proposal, and much more needs to be said (an understatement if ever there was one). However, as is perhaps obvious, the sentence used in the Romeo example has special features which can be misleading, so it will be useful to have a second one in play before I consider the proposal in more detail. For that I shall adopt one of Davidson's. Begin by supposing that someone has said:

(T) Leo Tolstoy is an infant.

Assuming that the background to this assertion makes it unproblematically literal—perhaps it is said on an appropriate date by one of Tolstoy's uncles—then we can understand ‘infant’ as making its usual linguistic contribution to (T). As Davidson puts it:

How is the infant Tolstoy like other infants? The answers comes pat: by exhibiting the property of infanthood, that is, leaving out some of the wind, by virtue of being an infant. … Tolstoy shares with other infants the fact that the predicate ‘is an infant’ applies to him; given the word ‘infant’ we have no trouble saying exactly how the infant Tolstoy resembles other infants. … Such similarity is natural and unsurprising to the extent that familiar ways of grouping objects are tied to the usual meanings of usual words. (Davidson 1984a: 247–8)

However, as used by a critic of the adult writer Tolstoy, as in the remark, ‘Tolstoy is a great moralizing infant’, that Davidson cites, (T) is metaphorical. Having noted this, Davidson mocks the idea that what we should now do is to find out what the set of infants, now including Tolstoy, have in common, thereby stretching the meaning of ‘infant’ to include this particular adult. The linguistic object ‘infant’ figures in a complex social practice, serving in that practice to label a property or properties we naturally discern. Nothing is gained by stretching that word-object so that it counts, either generally or on a specific occasion, as a label for a property-complex encompassing both toddlers and the adult writer Tolstoy. But no such stretching is required by the semantic descent story.

Understanding (T), when it is used metaphorically, begins with the fact that the word ‘infant’ applies to the usual suspects, namely the set of not yet grown up (p.96) human beings whose properties encourage us to group them together.4 No stretching of meaning, nothing special here. What happens next, however, is that, understanding this perfectly ordinary linguistic label, the hearer moves from it to an exemplar—to one of the set of things to which the predicate ‘infant’ applies—and then this exemplar serves as a qualifier of Tolstoy. It is as if, instead of coming out with (T), the critic said:

(1) Tolstoy is this …

while pointing to a 2-year-old child.

Here I expect questions to come crowding in. Since neither (R) nor (T) have explicit demonstratives in them, how are we to understand their relationship to sentences like (1)? Aside from issues surrounding (1), how does one choose an exemplar of a predicate? Does one actually have to have a specific infant in mind, or does any infant count? How can one be sure that the exemplar chosen will serve in the metaphorical setting? That is, how does one control the predicative use of an object? What happens if the exemplar doesn't happen to be, or is not thought to be, of the right sort for the metaphorical predication?

These are of course all reasonable questions, and they will be addressed at some point in this chapter. Basically, they cluster around two issues. On the one hand, there are questions about the movement, the descent, from words to objects—questions about the movement from ‘is the sun’ and ‘is an infant’ to such things as the sun and an infant. On the other, there are questions about the suitability of these objects to fulfil the qualificational role required by my account. The next three sections will concentrate primarily on the first of these issues. This is crucial because, unless I can defend the movement from words to predicative objects in metaphor, all the work put into the notion of qualification, and the work still to come, will be to little purpose. The remainder of the chapter will then concentrate on the issue of qualification itself.

3.2. Metaphorical Predication and Demonstration

First, let's look at:

(1) Tolstoy is this … (said while pointing to an infant),

and its supposed relationship to the original metaphorical utterance:

(T) Tolstoy is an infant.

My offering of (1) as a way of thinking about (T) is not intended as a proposal of strict equivalence. Still, it is not stretching things too far to think that the demonstrative (p.97) version captures something about the metaphorical sentence, and that is all I need. For I shall argue that when we understand the demonstrative sentence properly, it contains a clue about the metaphorical (T) which points to the correctness of the semantic descent proposal.

The issue of demonstratives used as in (1) came up in Chapter 2. There I was grappling with the question of whether the swatch and colour card cases—cases I called ‘sample series’—were genuine instances of qualification, or were instead merely cases in which some demonstrated object supplemented or filled out an otherwise linguistically articulated thought. This question arose because some might think that:

(2) My house is this blue (accompanied by pointing to a colour card),

is, as far as the demonstrative is concerned, not much different from:

(3) Put the ice in this.

This putative similarity is problematic because in (3) we have, as it were, full linguistic articulation: we recognize that complete practical understanding requires one to know the actual referent of ‘this’, something the context surely ought to provide, but there is nothing essentially linguistic missing. If (2) were similar, then this would ruin the point I was trying to make about the predicative-like role I claimed to have found in the demonstrated colour card.

My counter-argument required us to look at sentences such as:

(4) My house is light blue.

In (4) we have a structure parallel to that in (2), and exploiting this, I claimed that the predicative function of ‘light’ is in fact matched by the predicative function we must now allot to the demonstrated object in (2). Even though the colour card's predicative function is dependent on the linguistic predicate ‘blue’, it nonetheless has such a function. So, we cannot simply dismiss the demonstration in (2) as merely a contextual filling out in the way that seems natural enough in (3). The demonstrated item in (3) is important for a full understanding of the sentence, but doesn't itself have a linguistic function, while the colour card demonstrated in (2) does.

Whatever you think of this argument, I remind you of it again here because I shall not rely on it. Instead, I shall argue that we should distinguish sharply between what is going on in (1), and what is going on in both (2) and (3). When you come to see that the demonstration in (1) involves neither contextual filling out, nor even a subsidiary predicative role, my hope is that you will be prepared to see it as requiring the more radical treatment that comes with my semantic descent proposal.

Superficially, the utterance of (to remind you):

(1) Tolstoy is this … (while pointing to an infant),

resembles:

(2) My house is this blue (accompanied by pointing to a colour card).

(p.98) But, whereas the presence of ‘blue’ in (2) guides our use of the demonstrated object (the square on the colour card), what is going on in (1) doesn't fit this pattern. This difference is easier to see if we consider a more austere version of (2). Imagine someone uttering:

(2′) My house is this … (while pointing to a square on the colour card).

Though there is here no explicit use of a guiding predicate, one has no trouble at all in finishing this sentence with ‘blue’ or ‘colour’; indeed, (2′) seems to call out for some such completion. But try this doing something parallel with (1). Here are two possibilities:

(1′) Tolstoy is this infant,

(1″) Tolstoy is this human being.

Neither of these work in the way that the completions of (2′) do, nor, in fact, do they advance matters at all. In (2) and in the completed (2′), the demonstrative helps to secure a further narrowing of the general division of things into blue and non-blue, but this is palpably not what is going on in with the demonstrative in (1′) or (1″). Moreover, the latter sentences have the same bizarreness as the original metaphorical (T). Saying that Tolstoy is this (demonstrated) infant or this (demonstrated) human being is no improvement on saying (while demonstrating an infant) that Tolstoy is this. Nor would it help to insist that what is demonstrated in (1′) is the property of being an infant, rather than an infant itself. Partly this is because it is difficult to read (1′) as demonstrating a property, but mostly because properties don't help here.

Note first that (1′) is only superficially like (2). In the latter, one can be satisfied that what is demonstrated is the property of being a specific colour because, in the end, a property is precisely what (2) attributes to my house. But it makes no more sense—is no help with the metaphor—to say that Tolstoy possesses the property of being an infant than to say that he is an infant. Not only is it wrong to say that Tolstoy is one and the same as this (demonstrated infant), as apparently required by surface form of (1), it is no less wrong to say that he has the property of being an infant, or that he has the property of being this particular kind of human being, namely, the infant kind. (I will return below to further consider the role of properties in the demonstrative (1).)

It is important to be clear what is at issue here. As admitted above, I am not claiming that the use of the demonstrative in (1) to refer to a particular infant is simply fine. It isn't. The most straightforward way of reading (1) is bizarre, and some story must be told about what is really going on. But whatever that story is, we must have one. We cannot get away with thinking that (1) is fine because it demonstrates a property or properties, and is thus something like (2) or (2′). Moreover, the very fact that, as it stands, (1) is bizarre means that we cannot treat it as like (3); as like a case in which the demonstration has as its point the supplementation with something extra-linguistic of a perfectly intelligible linguistic construction. The demonstration in (1) is not straightforwardly like that in (3).

(p.99) Against this background, my suggestion is that (1) is best explained via semantic descent and qualification. As in ordinary sentences with demonstratives like (3), the demonstration in (1) aims squarely at a spatio-temporal particular. Though not without a certain strangeness, (1) requires us to supply an infant as the referent of ‘this’. In so far as (1) captures something of the original metaphorical (T), what we have here is semantic descent. (More on the relation between (T) and (1) below.)5 As noted, supplying a particular for ‘this’ leads to a certain strangeness: it invites the reading of the copula as an identity claim, as if we are saying, bizarrely, that Tolstoy is one and the same as some infant. However, by calling on qualification here we can overcome this temptation, and, at the same time, remove the strangeness of (1). For unlike the supplied referents in ordinary demonstrative sentences, the object demonstrated in (1) has, in addition to its being a particular in a context, a linguistic function: the infant answering to ‘this’ in (1) qualifies Tolstoy. In effect, the expression ‘is this’ in (1) is functionally a hybrid. It consists of a word ‘is’—understood as the predicate copula and not as the sign of identity—and the object answering to ‘this’. The copula and the object working together function as a predicate of Tolstoy.

Note the way the copula exerts some control over the qualification effected in (1). The object called upon by ‘this’ wordlessly exercises a predicational function, but this is partly because it is set in a linguistic structure typically marking monadic predication, namely, ‘is (a) …’. We are thus encouraged to understand the qualifying object as itself ‘monadic’. This helps with a problem that emerged in Chapter 2 in respect of qualification, namely that objects, in contrast to linguistic predicates, lack ‘slots’. In Chapter 2, I insisted that the absence of slots was not itself a reason to be suspicious of the idea of object-predication. What I claimed was that slots indicate, in a fully explicit way, what outside natural language can also be indicated by the circumstances within which the qualification takes place. But, on the proposal which finds a hybrid in (1), we can see how it is possible for the ‘adicity’ of a predication to remain a matter of words, while the predication itself is accomplished by objects and not words. (Of course, I have yet to consider examples in which objects function other than monadically. That will come mostly in the next chapter.)

The fact that the treatment of (1) starts with a fully linguistic construction (‘is this’) and ends up with a hybrid (‘is’ + object) suggests another kind of linguistic control operative in metaphorical sentences, but spelling out this suggestion will take a few paragraphs.

3.2.1. Qualification and linguistic control

The demonstrative ‘this’ invites us to pass with the least possible informational baggage from words to objects; all the work is (p.100) accomplished by the extra-linguistic circumstances or context within which ‘this’ is used. However, the original focus of my semantic descent proposal:

(T) Tolstoy is an infant,

did not involve demonstratives. As noted earlier, my interest in the demonstrative in (1) is that it captures something of the import of (T), though I never claimed any equivalence between (T) and (1), or between metaphorical utterances of this general form and sentences with demonstratives. For a start, ‘infant’ is not a demonstrative, nor could it be plausibly argued that this concept-expression contains a demonstrative element. Moreover, there are endlessly many metaphorical sentences in which explicit demonstration would be simply out of the question. Remember that ‘objects’, as I am using this notion, includes items such as events, situations, and states of affairs, some of which might well be non-actual in any straightforward empirical sense. (The problem of non-existent objects called on in metaphors will be discussed at the end of this section.) Still, within the context set by (T), it doesn't seem unreasonable to think that (1) at least approximates (T)'s message.

For a start, both sentences are problematic: on the surface, they are bizarrely false or perhaps just plain bizarre. Further, the source of this bizarreness is pretty much the same in both cases: the author of War and Peace is not an infant, nor is identical with this (demonstrated) infant. Still, if we take the hint that (1) offers, both sentences are made intelligible by my proposal. The sentence with the demonstrative ceases to be bizarre if we take the demonstrated object to be doing qualificational work. And the original sentence (T) likewise comes out alright if we take the concept-expression (‘infant’) as inviting semantic descent to an object that falls under this expression. For, though this descent is not accomplished demonstratively, the making of it nonetheless offers us a way of making sense of (T). Rather than taking it to claim infanthood of Tolstoy, we are free to take it as claiming instead that some infant qualifies him.

In this case of semantic descent, as in all others, we are invited to move from words to objects. Note though that the words from which we descend can play an important guiding role in qualification. Moreover, there is every reason to think that such guidance goes well beyond the fact that the object descended to must fall under the concept delineated by the words in the original sentence. This is the suggestion about linguistic control mentioned earlier.

Obviously enough, the infant used to qualify Tolstoy must be an infant. But there are other concept expressions that have infants in their sights. Thus, we might have been told, for example:

(5) Tolstoy is an early stage but independently viable human organism.

While it is true that any exemplar of this predicate expression is also an exemplar of ‘infant’, it is of course absurd to think that this sentence is just as good for metaphorical purposes as (T).

It might be thought that this is actually a worry for my semantic descent account of metaphor, but only by someone who had forgotten the lessons of the last chapter, (p.101) and is not paying attention to the central aim of this one. Qualification is not something that can be guaranteed to work just by wheeling in an object in the presence of some target subject. Context and circumstance are crucial to the intelligibility and aptness, as well as to the usefulness of any instance of qualification. Moreover, and crucially, when it comes to metaphor, we are not dealing with qualification on its own, but with qualification that arises from an encounter with an utterance or inscription in natural language. The words that figure in any such encounter are therefore as much a part of the context of the qualification as is any feature of the object itself. More specifically, it matters a great deal whether the words from which descent is made are as in (T) or as in (5). That is why the inappositeness of (5), so far from being a problem, is actually a pointer to an important positive feature of my account. The objects reached by semantic descent from the words in each of (T) and (5) may be the same, but the words themselves guide or control or prepare the ways in which we can be understood to use that object to qualify Tolstoy. This is not to say that these words guide us to different objects—that has already been made clear—nor do they encourage us to posit anything as problematic as ‘objects under descriptions’. Instead, in serving themselves as part of the context of utterance, the words exercise some control over the way the object got by descent comes to figure as a qualifier. Thus, infants are infantile, and this latter expression, while it can mean simply ‘pertaining to the early stage of human development’, offers more than a hint of the qualificational role that the object, the infant, is intended to play. In contrast, it is unlikely that the object got by descent from the words, ‘early stage but independently viable human organism’ would be taken to qualify the adult Tolstoy in the same way. This is not because the objects differ, but merely because the explanation of how the objects come to be used depend in part on the words which leads hearers to them.

An aside in two parts: first, the Tolstoy example makes the point about linguistic control seem weaker than it would be in realistically complex examples. As I keep saying, I will return to discuss the downside of this kind of simple subject-predicate example.

The second point is actually a sort of disclaimer. I am aware that, in making the point about linguistic control, and indeed in giving an exposition of the semantic descent account generally, it can sound as though I am propounding a psychological theory; as though what is in question is how we actually process metaphors. However, this should be seen merely as an artefact of the demands of exposition: making a point about how, from a theorist's point of view, one should account for a feature of an utterance tends to make for a great deal more circumlocution than writing as if one was adopting a hearer's point of view. More will be said about this later in this chapter.

3.2.2. Semantic descent and properties

Here let me consider one last issue to round off this section, an issue which requires us to return to the idea that properties are somehow involved in understanding (1) and hence (T). Earlier on, I noted just how (p.102) difficult it is to understand (1) as involving demonstrative reference to properties. We have no trouble in understanding the previously discussed:

(2′) My house is this … (pointing to a square on a colour card),

as demonstrating a property of the square, rather than the square itself. If for no other reason, this is shown by the naturalness of adding the property-invoking expression ‘colour’ to (2′). But there is no way to mimic this in the case of (1): adding the expressions ‘infant’ or ‘human being’, if it does anything, intensifies the bizarreness of (1) by emphasizing the particularity of the demonstrated object.6

There thus seems to be a real difference between the use of the demonstrative in (1) and in (2′); that is, in metaphorical and non-metaphorical contexts. My semantic descent proposal suggested a way in which we could leave untouched the reading of (1)—a reading on which a particular object is demonstrated—while yet rendering the utterance intelligible. This required us to see the process of qualification at work behind (below?) the surface reading of (1). Now someone might see this as an opportunity to insist that, if we were prepared to abandon this surface reading of (1), we could make it intelligible without appeal to anything as radical as qualification. In outline, this would work as follows. First treat (1) as actually saying something like:

(6) Tolstoy has some of the properties of this … (an infant is demonstrated)

Clearly enough, (6) takes liberties with (1), but if we allow them, doesn't (6) show how to render (1), and ultimately (T), intelligible? That is, isn't it perfectly sensible to explain the metaphor (T) as claiming, not that Tolstoy is in fact an infant, but that certain properties of infants apply to him?

Appeal to properties and the idea of similarity (understood as a sharing of properties) is perennial in discussions of metaphor: even before we got to metaphor, it surfaced in the discussion of qualification, and we will have reason to encounter it again. Here let me point out two interconnected things. First, if we are attentive, we should find (6) no more intelligible than the untampered with (1). Thinking carefully about the properties of infants—remembering all the while what they are properties of—it would be strange to think that Tolstoy has any of them. He does not totter around uncertainly, babble, smile gormlessly when funny faces are made at him, scream when put to bed or complain when denied access to potentially dangerous objects. Nor does he cling to his mother or drool. (I realize I am not painting a flattering picture, but the point here is to take seriously the idea that what is (p.103) in question are properties of infants.) To this the typical riposte would be: no one is claiming that Tolstoy has just the properties that an infant has, rather he has properties … well, like those of the infant—he is infant-like. Now, whatever else one thinks about this response, my second point is that this is no advance on our understanding of the metaphor which began the discussion.

Being told that Tolstoy is an infant is simply not the same as being told, for example, that he is a writer. (This is putting it mildly.) Some explanation must be offered of the former remark, an explanation that is not required of the latter. It sounds informative to say of the original (T) that its point is to attribute some of the properties of infants to Tolstoy, but this is an illusion. It may well be that the properties of infants can be thought of in some way—can be transformed—so that we can see them as applying to Tolstoy. I would scarcely deny this. But then the focus of our attention should be on the processes of transformation of properties, and we cannot, as some may be tempted, think of the claim in (6) as a finished job of work.

It will become clearer than it might be at present that some version of the appeal to properties is in direct competition with the semantic descent/qualification account. On my view, the understanding of (T) and (1) requires that we descend from words to an object (an infant), which is then used to qualify Tolstoy. The competition has it that (T) and (1) invoke properties of infants in the attempt to understand what is being said about Tolstoy. My view takes at face value the surface readings of the relevant sentences, but it requires one to accept my story about qualification; the other view requires us to put aside the surface reading, but it works with the familiar idea of a property. So far, it might be thought, too close to call. Not so.

Several things speak in favour of my account and against appeal to properties. However, engaged as I am in trying to make clear what is involved in my account, and especially in the semantic descent part of it, I shall only outline them; more criticism of similarity will follow in Chapters 4 and 5.

First, a property-invoking account should not be thought in genuine competition with mine unless it can cover the same range of examples. Given my present self-denying restriction to simple subject-predicate cases, the competition looks real enough: predicate expressions of the form ‘is an F’ lend themselves to talk of properties. But when it comes to syntactically more complex cases, it will be difficult even to formulate a property-invoking conception. Nor will the problem be just one of complexity. The richer examples to be considered in later chapters will involve certain phenomena of metaphor—phenomena such as that of deadness in metaphor and mixed metaphor—which are simply not amenable to the property treatment.

Secondly, even with respect to a simple case like (T), the field on which the two views play is far from level. On my account, the move from words to objects is crucial, and, though I haven't emphasized this, the same is actually true of the properties view. For what is in question in (T) is not the property which is characteristic of infants—the property of being one—but rather some property or properties of those things which fall under the concept-expression ‘infant’. In other words, there is a kind of implicit semantic descent, or something like it, involved even on the (p.104) supposedly competing account: you have to think of particular infants to get at their properties before you can even begin to think of how to transform these properties in ways appropriate to Tolstoy.

Thirdly, the unexplanatoriness of the properties story about metaphor is more fundamental than might appear on the surface. I have yet to give any real detail about the constraints needed to make qualification yield plausible explanations in this or that case of a metaphor. And someone might think that this lack of detail shows that my account and the properties-based one are, at least at this point, tied. But there is a difference: there is reason to think that, in respect of the properties-based account, the route to these details is blocked.

At bottom, the properties account treats (T) as claiming that Tolstoy and an infant share certain properties. As we have seen, it is not that specific features of infants are asserted to be features of Tolstoy. Rather, it is that there are ways of transforming features of infants so that they then can plausibly apply to Tolstoy. How does one go about transforming such a feature? An infant might scream when it doesn't get its way. Tolstoy doesn't. But, if we re-conceive this property of an infant, perhaps we can come up with something that does characterize Tolstoy. For example, suppose that, when certain of his purposes are frustrated, Tolstoy writes some bitter denunciation of the person or circumstance held responsible for his frustration. Can we not say that Tolstoy and the infant share the property of lashing out when thwarted? This kind of property transformation seems natural enough—though it is not always going to be as easy as this—yet there is a problem with it. When we transform properties in this way, so that they become bland enough to apply both to the infant and Tolstoy, we lose sight of what was metaphorical in (T). We have left behind properties specifically of infants, and it was these that made the metaphor apposite in the first place. This suggests a kind of catch-22 for the property account: if you don't leave the actual properties of infants behind, you cannot achieve the transformation required to render (T) intelligible. Yet, if you do leave them behind, you have somehow lost the metaphor. No such problem dogs my account: by insisting that the infant as such qualifies Tolstoy we keep the original metaphor firmly in the picture.

It would be easy to misunderstand my opposition to the property account. It is not that I think it wrong to say, in regard to (T), that Tolstoy shares certain properties with infants. I think this true, and I don't think that any writer on metaphor would deny it. Even Davidson can allow that someone can come to think, as a result of hearing (T), that relevant properties are shared. What I object to is thinking that all we need to do is to advert to some such sharing in order to account for the intelligibility of (T). I have outlined some of the reasons for this, and, as already noted, I will amplify and add to this list in due course.7

(p.105) Aside from this or that specific objection to the property account of metaphor, there is a more deep-seated reason for my thinking it fundamentally unappealing. The idea of qualification precedes—and at least partially explains—the ways in which we come to speak about concepts or properties. This was the burden of my remarks about the origins of categorization in the previous chapter. Putting these remarks together with my insistence that qualification is a crucial element in understanding metaphor, it should come as no surprise that I regard any appeal to properties as hopelessly too late. They have already been encountered as by-products of the account of qualification—the very notion at the centre of my account of metaphor—so they are unlikely to impress when they are re-encountered in property accounts of metaphor. For me, appealing to properties in order to explain metaphor is something like making introductions in a room full of people who know each other already. Your use of their names might be accurate, but whatever you are doing it is not effecting introductions.

3.3. Predicates and Exemplars

In this section, I should like to address some concerns about the examples that I have used. Initially, I shall try to be more explicit about the similarities and differences—especially as concerns semantic descent—between:

(R) Juliet is the sun,

(T) Tolstoy is an infant.

I introduced (T) because I was concerned that someone would think (R) loaded the dice too much in my favour. Romeo's remark, in having both the definite article and a proper name, suggests reference to a particular more directly than sentences of the more ‘standard’ metaphorical form typified by (T). By appealing to a metaphor with such an obvious referential device, it might have been thought that I was making it too easy for semantic descent. However, (R) is actually stranger than it might appear, and this is relevant to my account. Having been careful not to derive any undeserved support from any particular features of (R), I now want to show how these features in fact lend merited support to my account. (Long-held-over issues raised by the descent in cases like (T) will come in the second part of this section.)

3.3.1. Romeo's predication

I begin with some obvious observations. Sentences of the form:

_ is the …,

are by no means semantically uniform. Here are some ways of filling the gaps:

(7) Benjamin Franklin is the inventor of bifocals.

(8) Einstein is the brilliant scientist.

(9) Ernest is the most awful bore.

No doubt I have overlooked many other variations, but the above illustrates the range of possibilities I shall call on. The first asserts an identity by using two (p.106) referring expressions, though of course this is subject to issues about the referential status of definite descriptions that are not of concern here. The second has some claim to this same identity status, but it would be naïve to think that that was all there is to it. Unlike (7), it does not simply contain two ways of picking out individuals which are then asserted to be identical. In some way or other, the reference to ‘the brilliant scientist’ conspires with the copula to yield what is in effect a predicate of Einstein. The third sentence is also superficially similar to (7), but, like (8), doesn't seem to be simply an identity claim; it is impossible to avoid the feeling that the point of the copula here too is fundamentally predicational. Yet the superlative in the (9) adds an interesting extra element.

Suppose that, instead of (9), someone had asserted:

(10) Ernest is an awful bore.

Ernest might not think so, but this is a weaker condemnation than (9). Moreover, this is pretty much what one would expect: changing the definite to the indefinite article seems bound to have some weakening effect. However, consider the claim:

(11) Ernest is a most awful bore.

Though containing the indefinite article, those I have asked find this sentence to have pretty much the same strength as the original (9). There is, though, something strange about it: while my informants found (11) perfectly idiomatic, the superlative ‘most’ and the indefinite article do not really go together. This is perhaps clearer in a case where the superlative form of an adjective does not rely on outside help from ‘most’. For example, consider the distinctly odd:

*(12) Ernest is a fastest runner.

What this suggests is that there is some pressure to write ‘the’ in sentences like (9), even though they are intelligible with the indefinite article, and make the same point as their definite article versions. It is as if our desire to point out the extremity of Ernest's boringness leads us to the hyperbole of ‘most’, and this in turn puts pressure on us—grammatically though not necessarily semantically—to use a definite article.

The various points made about sentences (7) to (9) will come in handy in dealing with Romeo's (R). However, before I come to that, note a respect in which his sentence differs from any of them. The definite article in:

(R) Juliet is the sun,

is semantically (though certainly not grammatically) redundant: the expression ‘the sun’ is one of those cases where the definite article, and what follows it, form a semantically unitary, though typographically complex, proper name. There might be some resistance to this observation because of the tendency of non-astronomers to use ‘sun’ to mean ‘star similar to the sun in having orbiting planets’, or even sometimes just ‘star’. But, leaving aside the fact that Shakespeare's sentence would not have been understood this way, and the fact that it is strictly incorrect astronomical (p.107) usage, we can avoid arid controversy by the simple expedient of capitalizing. (In fact, some writers do this anyway.) Thus, from now on think of Romeo as having said that Juliet is the Sun; surely, in this sentence the definite article is redundant in the way described.

Now, given that (R) involves reference to a unique, actually existing particular, it would seem that (7) is the sentence we should look to; we should see the copula in (R) as that of identity. However, this is most certainly not how this sentence is taken, and this is somewhat mysterious. Many writers have taken Romeo's sentence as a good example of a metaphor, not least because it has a kind of vividness absent in the usual ‘Harry is a fox, wolf, tiger …’ sorts of case so common in philosophical treatments of metaphor. But when this sentence is held up as a reasonable example of a metaphor, the fact that it looks like an identity claim is not even noticed. Though I haven't done a head count, I cannot think of any writer who treats (R) as other than a better (more vivid) case of a subject-predicate metaphor. In an effort to shed some light on this, let me return now to sentence (8).

As noted above, the expression ‘the brilliant scientist’ is not understood in (8) as simply referring to a specific individual who turns out to be Einstein. Only someone impervious to the nuances of language would hear (8) as just like (7). Instead, we understand (8) as somehow using the superficially referential expression, ‘the brilliant scientist’, in a more purely predicative role.8 How might we cash the vague ‘somehow’? Well, suppose the expression ‘the brilliant scientist’ directs our thinking towards an exemplar, towards an individual, not necessarily a historical figure, who is quintessentially a brilliant scientist. And suppose further that, in focusing on this exemplar, we come to think about Einstein in a particular, or even new, light. (I am imagining that (8) is used in a context where it is informative and not merely emphatic. Think of someone saying it to students in a high-school physics class who are perhaps too young to have much of an idea of who Einstein was and is.) What we have here is of course a form of semantic descent followed by a process like that of qualification, and though there might be other stories that one could tell about what is going on in (8), I can't think of any that so naturally explain how what looks like straightforward reference to an individual could be turned into something predicational.9 (Note though, that I stop short here of saying that the exemplar here works in precisely the way that it does in metaphors. More on this important issue nearer the end of this section.)

The minor mystery about Romeo's remark that was described earlier disappears completely if one sees it as like (8): the descriptive name—‘the Sun’—does indeed refer, but the sentence is not naturally understood as like the claim that (p.108) Benjamin Franklin is the inventor of bifocals. Instead, we take the referent as itself having a predicational (or better, given that an object does the work, qualificational) effect, and that is why (R) is assimilated to the usual range of subject-predicate examples of metaphor. Someone could I suppose insist that (R) does assert an identity whose patent absurdity alerts us to its metaphorical character. But this is just not the way it tends to be taken. Romeo is not informing us that the world contains an object which happens both to be Juliet and the Sun—he is not identifying her with the Sun—though he is certainly using the Sun to tell us something about her. Any patent absurdity in his remark belongs with the ‘usual’ metaphorical absurdity we find in claims about a man Harry that he is, for example, a fox, or a wolf or a lion. (Note, though, that I am not here signing myself up to the idea that patent absurdity or falsehood is necessary to the identification of metaphor.)

Suppose instead of (R) that Romeo had said:

(R′) Juliet is a Sun.

Is (R′) even intelligible? I ask this because, given that there is only one thing that answers to ‘Sun’, the existential ‘a’ might be thought inappropriate. However, a little informal canvassing has convinced me that (R′) is not only intelligible, it is in fact acceptable, indeed it is almost as strong as Romeo's original remark. In fact, the relationship between (R) and (R′) is very much like that between (9) and its existential variant. So that you can take them all in, they are:

(R) Juliet is the Sun.

(R′) Juliet is a Sun.

(9) Ernest is the most awful bore.

(11) Ernest is a most awful bore.

The slight difference in strength between the first and second of each pair (and the hint of oddity in the second sentence in each pair) has precisely the same cause. ‘Most’ in the second pair implies uniqueness, as does the capitalized ‘Sun’, so it can seem odd to team them up with the uncommitedly general ‘a’. Yet when we find ‘the’ instead of ‘a’, this doesn't itself inform us of the uniqueness of ‘most’ and ‘Sun’—we knew of this uniqueness already, so in both cases ‘the’ is strictly redundant—but it is still reasonable to think of the uniqueness as reinforced. This is perhaps the reason why the second of each pair is heard by some as slightly weaker than the first, though others find them merely stylistic variants.

I have spent a lot of time on the nuances of the definite and indefinite articles (in a single natural language) because I believe them to show something important about subject-predicate metaphors. More particularly, what they show seems to me to constitute further support for the semantic descent account. Here is how the story goes.

The sentence:

(13) Harry is an accountant,

(p.109) seems to be syntactically and semantically of the same form as:

(14) Harry is a fox.

But there is more going on in (14) than is usually recognized. In whatever way we manage it, let us suppose that (14), on some specific occasion of its use, is identified as metaphorical. (As I keep saying, my account of metaphor does not take any line of the issue of identification, but we can surely agree that, even if it is not particularly vivid, (14) is a metaphor.) Having identified (14) as metaphorical, we certainly realize that, whatever else is going on, we will not be able to explain adequately the linguistic act effected in its utterance by treating it as just like (13). However, as a prelude to any heavy-duty theorizing (including my own), consider:

(14′) Harry is the fox.

This sentence might be less conventional, but it is surely just as good as (14) for conveying the metaphor. Now, as per previous discussion, there are two ways to read the copula in (14′). On the one hand, we can see it as inviting an identity between Harry and some specific fox, rather as in ‘Benjamin Franklin is the inventor of bifocals’. On the other, we can see it as predicational, rather as in ‘Einstein is the brilliant scientist’ or ‘Ernest is the most awful bore’. Given the implausibility of the first reading, the second is certainly indicated, and quite predictably it clears the way for my preferred account. That is, given that the expression ‘the fox’ is most naturally taken as referring to an exemplar—a specific yet typical instance—of the predicate expression ‘fox’, we have to find some way to combine this reference to an exemplar with the predicational aspirations of the ‘is’ in (14′). And this is precisely a situation that semantic descent and qualification were designed to accommodate. In referring to an exemplar, we semantically descend from the words ‘the fox’ to a relevant object, while preserving the predicational aspirations of ‘is’ by treating this object as itself a predicate—as a qualifier of Harry. In the way described earlier, we are forced to recognize that (14′) contains a hybrid predicate, one composed of the word ‘is’ and a non-word object. (Remember too that we cannot get away with seeing ‘the fox’ as contributing the straightforward property of being a fox to the sentence, since this is certainly not something (14′) attributes to Harry.)

What about the humdrum:

(13) Harry is an accountant?

Everything about the argument of the previous paragraph depends on the fact that meaning, and metaphoricality, are preserved in the shift from the indefinite article version in (14) to the definite article version in (14′). If the same shift can be as easily and conservatively effected in the case of the standard subject-predicate (13), this would create problems. For surely it is implausible to see semantic descent and qualification at work everywhere. As it happens, though, this is not a problem I will have to face, as the shift in (13) leads to:

(13′) Harry is the accountant,

(p.110) and the most natural interpretation of this calls on the identity reading of the copula, not the predicational one; it is thus quite different from (13).10 Nor can this difference be attributed to the fact that ‘fox’ is a predicate expression for a natural kind, whereas ‘accountant’ is not. (I mention this because someone might think that reference to an exemplar is typical of natural kind terms, and only of them.) For a start, ‘infant’ is not such a term, though the article shift also works here. Moreover, a little reflection shows that the article shift works just as well when the terms in play are least like natural kinds. Try it with the metaphorical: ‘Ruth is a (the) bulldozer’; and the literal: ‘The yellow vehicle parked in the field is a (the) bulldozer.’)

Though simple subject-predicate metaphors are a highly restricted category, the fact is that they differ in the way just described from simple but literal subject-predicate sentences, and this difference is congenial to the semantic descent account. In effect, ‘standard’ metaphorical subject-predicate sentences like (T) or (14) have more in common with the non-standard (R) than might at first have seemed plausible.

In the next chapter, I will consider metaphors of widely different syntactical forms, showing in fact how, in dealing with these complications, my account gains even more support. Still, it is worth having spent some time on the subject-predicate form because, aside from its prevalence in philosophical discussions, it can seem least suited to my account. This is because my account requires us to find some object to which we are referentially directed by some plausible understanding of the words used in the metaphor, but the indefinite article 1 noun structure typical of subject-predicate cases does not immediately suggest reference to some object. I earlier attempted to get around this by calling on a demonstrative version of the Tolstoy example, and this might have done the trick for some readers. But parity between:

(14) Harry is a fox,

and:

(14′) Harry is the fox,

is intended to offer further support. I claim that, however much subject-predicate metaphors resemble ordinary subject-predicate sentences—sentences innocent of object-reference—they are not. They more closely resemble the sentences:

(8) Einstein is the brilliant scientist,

(9) Ernest is the most awful bore,

(p.111) than they do non-metaphorical sentences of the form ‘A is a B’. And, while (8) and (9) are not themselves metaphorical, these sentences contain descriptive phrases which are at least candidates for reference.

Nonetheless, they are not metaphors, so does the semantic descent account apply to them and, if so, does this invite further worries about the promiscuity of that account? The answers to these questions support rather than threaten my account. On the one hand, it does seem as if something like semantic descent is a part of our understanding of these sentences. Each invites us to think of an object—the brilliant scientist/the most awful bore—and thus there is the suggestion of a move from words to objects that is characteristic of semantic descent.

On the other hand, the role of the object got by this semantic descent is somewhat different from the relevant counterpart in metaphorical contexts. Thus, having thought of an object answering to the description, ‘the brilliant scientist’, and regarding this object as a qualifier of Einstein, we can nonetheless spell this out in a way not available in the metaphorical case. Thus, we can treat features of the envisaged brilliant scientist as attributable directly to Einstein: this is part of what makes the sentence ‘Einstein is the brilliant scientist’ a literal predication. No such possibility exists in the case of Juliet and the sun, or Tolstoy and some particular infant. As already noted, there is a temptation to think that the property-sharing story can explain these metaphors, but, as I have argued, it doesn't, though, as the Einstein example suggests, property-sharing might well be all we need in non-metaphorical cases of descent and qualification. If the brilliant scientist is someone who is careless about practical matters, but able to make huge imaginative leaps in the attempt to explain the natural world, then we understand (8) as claiming that these features apply to Einstein. Indeed, any property of the brilliant scientist is intelligible when applied to Einstein, though of course there can be arguments about which ones are most salient or true. In contrast, think of features of the sun and Juliet: the sun is a nuclear furnace, it is responsible for the earth's warmth and light, it is used (was used) to measure time, etc. These can scarcely be attributed to Juliet without calling on some story of property transformation which is not easy to tell, and which would in any case undercut the explanatory pretensions of the appeal to property-sharing. Moreover, as I have earlier argued, in telling a story about how properties of the sun might be ‘sanitized’ sufficiently to be attributable to Juliet, we would have lost sight of the original formulation that we were trying to explain.

An interim conclusion: the sentence Romeo used is generally taken as a subject-predicate metaphor, but someone might think that, since Romeo's sentence contains a referential expression in its predicate, my use of this example unfairly softens one up for my semantic descent account. In fact, it is much more complicated: for a start Romeo's remark is not strictly of the subject-predicate form. However, when we look at this sentence more closely, and compare it with sentences like (8) and (9) above, it is possible to expect a reader to be genuinely (and fairly) moved in my direction. Admittedly, I am only basing this conclusion on the evidence of some pretty hair-splitting comments on a range of English-language sentences. But this evidence is (p.112) better than nothing, and I mean you to add this to what I have given so far and also to what is yet to come.11

3.3.2. Tolstoy and other infants

Most of the effort of section 3.3.1 was expended on the Romeo example, but there are more things to be said about the one involving Tolstoy. As I have several times acknowledged, there is bound to be a certain resistance to the idea that (T) involves semantic descent to a determinate infant, or even to an exemplar. Does ‘infant’ really involve reference to a determinate individual? Do we really have a particular infant in mind? There are several strands to my response to these questions.

On the one hand, it is important to remember that my account of metaphor is not intended as an account of how speakers and hearers process metaphors. Though it is often convenient in my exposition to talk about what goes on when someone encounters a metaphor, semantic descent is not in the end meant as a description of any such actual ratiocination. I am thus not claiming that we hear (T) and mentally work out which, if any, determinate infant is in question. Instead, the semantic descent account attempts to characterize metaphors in such a way as to make their intelligibility possible: the thought is: were we to allow descent from ‘infant’ to a determinate individual, and were we to imagine this individual taking on a role usually played by a predicate—a role I called ‘qualification’—then we could make sense of (T).

This distinction in perspectives should be familiar from the literature about theories of meaning for non-metaphorical utterances. On the model I think we should favour, characterizing what is generally thought of as the meaning of a word or phrase is part of the project of making sense of utterances involving that word or phrase. It is not a psychological account of what speakers and hearers actually know, though it is an account which suggests that, if they knew the relevant meanings, we would find their interchanges intelligible. (Actual speakers and hearers might process utterances in ways that do not directly require them to know the meanings described by the theory; the theory simply shows how the intelligibility of utterances comes to be possible.12)

(p.113) Long experience of trying to explain this distinction has convinced me just how difficult it is to grasp, never mind defend, but if you are willing to adopt the perspective on offer, it should lessen somewhat the oddity of treating ‘infant’ in (T) as calling on a determinate infant or exemplar. It is not that I am insisting that a hearer struggles to bring this individual to mind as part of comprehending (T). Hence, it is no objection to my account that you cannot find yourself engaging in any such mental activity when you hear this sentence. The claim is rather that, in adopting a theorist's point of view, assigning an individual to ‘infant’ is a way to make (T) intelligible.

All that said, honesty requires me to admit that, aside from its being controversial, the distinction of perspectives is not by itself enough to allay worries about the descent from ‘infant’ to an infant. So, a second defensive strand is called for, and it is at this point that I need to be explicit about the ways in which simple examples like (T) can mislead us into having unnecessary worries about semantic descent.

What is obvious about (T), and about virtually all similarly simple subject-predicate metaphors, is that they tend to be, at the least, tired, or, even more often, dead. A useful hint that they are like this comes from the fact that many dictionaries actually list a meaning for ‘infant’, which would in fact extend to the adult Tolstoy, and all dictionaries include such a meaning for the closely related ‘infantile’. Now the very fact that simple subject-predicate metaphors are dead, or close to it, does not by itself help me out in respect of descent. If (T) is a dead metaphor, and many think that this means it is in some sense no longer a metaphor, why did I use it? And even if it is only tired, how does this affect the issue of semantic descent?

As these questions show, appealing to the energetic status of (T) to allay worries about semantic descent would seem to require deeper investigation of the general consequences of that status, and that is something not attempted until Chapter 4. But even without taking on the whole of this task here, there is a way forward.

Begin by assuming or even just suspecting simply this much: precisely because (T) and its simple subject-predicate cousins tend to be tired or even dead, less is required from any account of metaphor for their intelligibility. In respect of my account, this means that there is less pressure to use the full resources of semantic descent to make (T) intelligible; what is known about ‘infant’, something shown in any reasonable dictionary, removes some of the need to descend to a determinate object, which then serves as a qualifier, in order to make the metaphor intelligible. This is of course only an assumption, but it gains support, even in advance of a full appreciation of the phenomenon of dead metaphor, by imagining ways to inject some energy into tired or dead metaphors like (T). Instead of the simplest subject-predicate form that we have in (T), think of:

Tolstoy is a infant who has just had his favourite toy taken away,

or:

Tolstoy is an infant who cannot get the attention of his parents.

(p.114) I am not saying that these are metaphors worthy of great literature, but they are certainly more vivid than the original. Crucially, moreover, they are also cases in which we are less inclined to think of what is in the dictionary, and more open to entertaining the idea that a particular infant figures in our assessment of Tolstoy. And this is precisely what one would expect, given the way the semantic descent account is meant to work.

The above consideration is most certainly not intended as direct support for the semantic descent account; it is intended only to defuse an objection which is, I think, an artefact of the simplest kind of subject-predicate example. Useful though they are for allowing me to sketch my account with a minimum of fuss, their very simplicity can be problematic. A full appreciation of these problems will have to wait on the discussions in Chapter 4 both of complexity and of the phenomenon of dead metaphor. Still, the fact that one can breathe life into a tired subject-predicate metaphor precisely by making the predicate more likely to fix on some individual should make more plausible the idea that a determinate object plays a role in the full account of its intelligibility.

Given the above considerations focusing attention on the way objects come into an understanding of metaphor, I can no longer ignore a problem that has been ticking away in the background. I begin with some comments about actuality.

My account might seem to depend on there always being some actual object that is in the extension of, or that serves as the exemplar of, the relevant predicate expression in a metaphor. But this is simply not true. The fact that all the cases of qualification in Chapter 2 involved actually existing items is perhaps partly responsible for this misleading impression, as is the unhelpful fact that the Romeo example uses the sun, an actual object if any is. However, it shouldn't be surprising that all pre-linguistic, pre-metaphorical cases of qualification involve actual objects, since the only kind of confrontation one could have with an object in these cases is perceptual. However, in spite of the Romeo example, this is not typically the case. Metaphor is something we do with words, and it has been my contention that we can best understand what is done when we take some of the words in a metaphor as confronting us with an object. But this kind of confrontation is quite different from that involved in perception. Words like ‘infant’ or, in a more plausibly vivid case, ‘infant who has just had his favourite toy taken away’ apply to things, have extensions, and the first part of the task of making metaphors intelligible requires us to conceive of some determinate infant which lies in some such extension. Conceiving, perhaps even imagining, some such infant is what is required by semantic descent, but there is no requirement here that the object be actual, that it be something we could see, touch, or pass from hand to hand.13

(p.115) It is precisely at this point that the worry alluded to above can no longer remain in the background, and it can be summed up in a simple question: what about unicorns? Philosophers who read my comments about conceiving of a determinate member of the extension of some predicate will instantly wonder how there could be semantic descent when a predicate is, as is said, empty. There are infants, and it might be allowed that metaphor requires us to conceive of being confronted by one of them in the sense required, but what about a metaphor built around, for example, words like ‘unicorn’? My answer to this question can be fairly brief.

The kind of empty predicates that are in question here belong to fiction in a broad sense, and fiction creates problems in all sorts of contexts; there is nothing special about the problems it creates for the semantic descent account. Moreover, though there is no consensus about the right one, there are perfectly reasonable strategies for dealing with fiction. Still, while I don't think there is a pressing need, a few further words about how one strategy could work might help to defuse the worry about unicorns.

From my point of view, the most promising strategy in respect of fiction is the one calling on the notion of pretence. Familiarly, this strategy allows that, in using a sentence with a fictional name or predicate, we do not really refer to fictional entities, but rather pretend that we are referring to real ones. A huge amount has been said about this strategy, and no doubt even this summary description of it could be faulted. But, even without looking too closely at details, its appeal in the context of my account of metaphor is obvious. Were a helpful friend to describe an adversary you find intimidating this way:

Walter is the hound of the Baskervilles,

it would be perfectly reasonable to find this metaphor intelligible by semantically descending to the poor dog who in the story is got up to be frightening, but is in fact harmless, and seeing this creature as qualifying Walter. Even granted that, harmless or not, there happens to be no hound of the Baskervilles, there is no special problem in pretending or imagining that he does exist, and therefore in taking this pretend individual to characterize Walter.

There is obviously a lot more to be said about the scope of pretence in claims like that about Walter; the use of fictional entities in metaphors will almost certainly complicate any pretence account of fiction. But I hope that what I have said is enough to convince you that fictional objects, when we pretend them to be real, can play pretty much the same role in semantic descent and qualification as any other object. Nor should this be surprising, given that we need to exercise imagination in semantic descent even when the relevant predicates have non-empty extensions.14

(p.116) 3.4. More on Semantic Descent (and Some Notation)

The notion of semantic descent was introduced by reference to the well-known Quinean idea of semantic ascent. In this section, I shall say something further about this, introducing along the way some notation that will prove useful.

In a typical case of semantic ascent, one takes an ordinary subject-predicate sentence:

(15) The sky is blue,

and treats that very sentence as a subject of comment via the truth predicate, namely:

(16) ‘The sky is blue’ is true.

This counts as ascent simply because (16) is metalinguistic, but it counts as semantic ascent because, in Quine's words: ‘The truth predicate is a reminder that, despite a technical ascent to talk of sentences, our eye is on the world’ (Quine 1970: 12).

Quine did not envisage the kind of case I have been describing, but nothing he says requires all cases of semantic ascent/descent to have the feature he associates with ‘true’, namely that the higher level sentence says the same thing as the lower-level sentence.15 One could regard preservation of content as a special case, appropriate only to ascent/descent using the truth predicate. Clearly, in the cases of semantic descent that I have described, sameness of saying is not preserved: taken literally, Romeo's (R), and the claim about Tolstoy in (T), are certainly not equivalent to their semantically descended counterparts. That said, there is still every reason to count my suggestion as involving genuine semantic descent. Our encounter with the words ‘sun’ and ‘infant’, together with our recognition of their context as metaphorical, leads us to employ the relevant non-linguistic objects in new, ‘hybrid’ predicates which, for convenience, could be displayed using the following notation:

(Rm) Juliet is the ↓sun↓,

(Tm) Tolstoy is an ↓infant↓.

Note: like quotation used in semantic ascent, the ‘↓’ marker of descent works in pairs, but there are differences and these will be discussed.

The result in each case is a linguistic structure in which one element, the predicational part, aims to give information about the other element, the subject part. (Of course, in each case, the predicational part is a hybrid, consisting of some words and an object. But that doesn't affect the present point.) Even though the predicates in (Rm) and (Tm) are not the same predicates as those in (R) and (T), and do not therefore convey the same information, they share a subject matter with their undescended versions. Thus (Rm) like (R) is about Juliet, and (Tm) like (T) is about Tolstoy. The descent in each case is made with our eyes still firmly on the (p.117) world—indeed on the same bit of the world—so Quine's condition for the semanticity of ascent/descent is still fulfilled.16

The down-arrow notation displays semantic descent in the various examples, but more must be said about the rationale behind this notation before I can justifiably put it to further use.

When someone utters a sentence in a particular context, something about the words used, or features of the utterer's performance, or the context itself, or some combination of these, alerts us to the fact that the utterance involves a metaphor. Exactly how we manage this identification, as I have noted several times, seems to me an often messy empirical matter, best decided by (perhaps) psycholinguists, and I have concentrated instead on the philosophical task of giving an account of what is identified. Given this division of labour between the empirical and philosophical, the down-arrow notation should be seen as a tool for the philosophical theorist. It is intended to record—from the perspective of my particular theoretical account—what it is to be metaphorical, and should not be taken as a marker whose independent identification triggers the judgement that an utterance is metaphorical.

This warning is necessary, not least because down-arrows were compared earlier to quotation marks, and the latter are of course not merely creatures of theory.17 But, so long as we keep this warning firmly in mind, the comparison between down-arrows and quotation is suggestive of something important. Quotation marks are a linguistic device that allows us to talk about the bit of language they enclose. If we remove them we get another linguistic device that allows us to talk about some appropriate bit of the world. Extending the process by using down-arrows around a predicative expression, we create yet another device, but one which now allows us to use one bit of the world to talk about some other bit.

Someone might hesitate to follow me here, insisting that ‘semantic’ and ‘predicative’ apply only to practices that involve words. The thought would be: when down-arrows are used, the result is an object, not words; that is, not something that we can use to talk about the world.

More or less this very issue featured prominently in the previous chapter: predication, being understood as essentially word-involving, I had to coin ‘qualification’ as the wordless undertaking of the predicational function. In so doing, I had in mind the idea that qualification should be counted the broader of the two notions, and that predication was therefore a species of qualification. The model here is reference: we manage to refer with objects, gestures, sounds, thought constituents, as well as words, and we don't even have a special term for reference by words. In contrast—and, as I argued, unreasonably—we think of predication as essentially something (p.118) achieved by using words, and we neither have a term for the more general activity, nor, partly for this reason, do we give it its due.

On the picture of qualification as the superordinate notion, there should be no problem in regarding the objects got by descent (in metaphors) as fulfilling a semantic function. In some cases, we use words to qualify items in the world (i.e. we use predicates); in others, descending from words to objects, we use these objects without the help of words to qualify items in the world. On the assumption that predicate-style qualification is a way of talking about something—a way of engaging in a semantical enterprise—there should be no problem in regarding object-style qualification in this same way.

That said, I could imagine someone finding this terminological proposal a little too radical. It is one thing to regard qualification as something like predication—close enough to find the parallel interesting—and another to treat qualification as the superordinate term of the pair. But all that I need in the present context, indeed, all I need for my account of metaphor, is the minimal concession that qualification is something like predication. Instead, therefore, of treating qualification as the superordinate term, we could think of it merely as a sort of precursor to fully semantic predication, that is, as ‘proto-semantic’ or ‘proto-predicative’. Not only does this not take anything substantial away from my account of metaphor, this way of putting the matter has useful resonances. After all, in so far as metaphor calls on a process which gives us insight into our ordinary notion of predication, it becomes itself implicated in the foundations of language. (I have already mentioned the foundational significance of metaphor, and it will figure more prominently in Chapter 4. At an appropriate point later in this chapter, I will return to the issue of the semanticity of hybrid or proto-predicates, since I certainly do not think that the remarks in this paragraph settle the issue.)

The idea of semantic descent, the proto-predicates it creates, and the down-arrow notation for displaying these predicates, are none of them everyday items, either in or outside philosophy. But there is a sort of language-like construction in common use whose familiarity might help here, at least expositionally. Think about what is going on in the ubiquitous:

I ♡_.

For all practical purposes, the heart-shape in this structure has now become synonymous with the word ‘loves’, so I cannot claim that it offers us an actual case of semantic descent. However, one could imagine a time when there was no conventional link between the symbol and the word. Thinking back to that imagined time, we can imagine further the heart-shape being invoked as a device for referring to the heart, an object conceived of as, roughly, the seat of our affections. In my terms, one could think of the heart-shape as combining the functions of the down-arrows and any words they enclose. Instead of ↓sun↓, we have ♡: the first picks out the object in the sky, the second the organ in our breasts. With this background, one could then finally imagine someone constructing the structure shown above. In this structure, the heart-shape picks out an object which itself then functions as a hybrid (p.119) or proto-predicate, one which qualifies the relationship between the speaker and whatever fills the blank.

Note that the heart-shape has a fixed effect, or one might say ‘meaning’: in every context, it picks out the heart. In contrast, we can think of the down-arrow notation as a functional device for turning almost any ordinary predicate into a proto-predicate. It offers us a way of concocting any number of symbols whose effects, in a given context, are something like that of the heart-shape. Rather than having recourse to the commercial artist's ability to think of expressive shapes, the least talented amongst us can turn words into such ‘shapes’. The down-arrow notation in effect marks the capacity of metaphor to harness the expressive power of language for uses other than that of ordinary predication. Also, the heart-shape example hints at something that I shall explore in the next chapter: serving as a two-place proto-predicate, a verb, it offers the merest hint that the semantic descent account, unlike many others, can cope with metaphors of widely differing syntactic forms.

3.5. The Route from Here

In the previous chapter, I introduced and defended the idea that objects can sometimes take on roles typically thought of as linguistic. In this one, I have set myself the task of using that idea in the semantic descent account of metaphor. What this account requires is, first, a movement from the words used in metaphors to objects, and, second, the use of these objects as qualificational, perhaps proto-linguistic, devices.

So far in this chapter, I have concentrated on the first of these requirements, that is, on offering some defence of the idea that the word-to-object movement in metaphor is something available and plausible. Further defence will of course have to be given. What I haven't yet done is to consider the issues raised by the second aspect of the semantic descent account. That is, I have yet to consider how qualification works in specific metaphors. In particular, I have yet to say anything in detail about the kinds of constraints that might make it reasonable and informative to maintain, for example, that the sun qualifies Juliet.

Saying the right things about both linguistic and non-linguistic context will be crucial here, but this is no surprise: context is bound to be a large part of anyone's story about metaphor. For instance, almost every writer who uses the example takes into account the fact that Romeo's assertion occurs in a larger linguistic context, one which gives useful, even necessary, guidance to interpreters. Of course, how this guidance is brought to bear will depend on one's account of metaphor. In my account, linguistic context works in part by giving us some idea of the kind of syntactical role an object is to play, and in part by hinting at its qualificational role. The sun might well work in one way in (R), set as it is in Romeo's extended monologue about a love-object, Juliet, and in another way, when this same object is used metaphorically in a story about the warrior Achilles.18

(p.120) No less, and often more, important is non-linguistic context. As with linguistic context, the function of this kind of context crucially depends on the account of metaphor in question. Though more will be said, it should be obvious that, on my account, the focus of non-linguistic context is, in each case, the object of metaphor got by semantic descent. That there is this focus for the information we think of as contextual is no small part of the reason to look with favour on my account. By insisting that there is such a thing as an object of metaphor, and that the purpose of this object is in fact predicative, we give contextual information something specific to work on and with. Accounts which merely point us in the direction of context, and suggest that it somehow helps us understand what is being got at in a metaphor, fall down badly in this respect.

Relevant contextual knowledge of the objects of metaphor can take many forms: it might be a matter simply of commonsensical knowledge, called on by the setting of the metaphor; or it might be a matter of conventional or even, to use a fancy term, cultural knowledge. After all, the sun, along with many of the other objects we call on in metaphor, can be described as having cultural significance (or significances)—a way of putting the matter which suits my purposes, given the resonance amongst significance, sense, and the wonderfully dextrous meaning.

Later in this chapter, I shall consider these matters further, though the nature and relevance of context to my account can only be fully appreciated when more complex examples of metaphor are in play in Chapter 4. However, aside from filling in details of semantic descent and qualification, there is a further important bit of unfinished business. In Chapter 1, I argued that each of the most prominent accounts of metaphor in the literature fail to accommodate at least one of the truths I described there as non-negotiable. It is thus crucial to show that the semantic descent account does better. For aside from worries about this or that aspect of the account, unless it manages the three truths about metaphor, it is no advance on those currently in the field.

It might seem best to take on the tasks in the order described above: first fill in some more details, and then show how the semantic descent account deals with all of the truths canvassed in Chapter 1. However, I have decided that the reverse order is actually better. Enough has been said about semantic descent (in this chapter), and the practice of qualification that grounds it (in this and the previous one), to make my overall account of metaphor clear, even if it needs further elucidation to be fully convincing. And it seems to me the best hope of convincing anyone is, first, to consider how an account shaped like mine negotiates the three basic truths, and then, once it is clear enough that some such account has a real future, to fill in the further details. Nor is this simply a tactic: some of the further detailed questions about semantic descent and qualification will be sharpened, even answered, by the discussion of the non-negotiable truths, and this will make the final sections of this chapter more rewarding.

What follows, then, in the next three sections will be a consideration of how my account manages to accommodate the truths about metaphor, followed in another section by further material on semantic descent and qualification.

(p.121) 3.6. Semantic Descent and Truth

The point about assertion and truth is easily accommodated. So long as one is careful not to equivocate, semantic descent can explain why it is perfectly in order to question, contradict, or assent to metaphorical utterances, and nothing in this explanation requires us either to eschew genuine metaphorical content, or to locate the focus of these truth-evaluating activities in something other than that content. Insensitive though it may be, an interlocutor may counter Romeo's claim, without changing the subject, by telling him:

No, Juliet is not the sun.

Unlike Davidson, I do not regard this remark as obviously true. For, as intended by the interlocutor, we must see this counter-claim as using the proto-predicate demanded by semantic descent, namely, ‘is the ↓sun↓’; treating the word ‘sun’ here as part of an ordinary linguistic predicate would be to equivocate, thereby turning the interlocutor's remark into a bad joke.

The fact that the semantic descent account preserves the intuition about metaphorical truth should be unsurprising, since, as has been noted, it comes under the head ‘Content Sufficient’. It belongs with the proposals of Beardsley, Black, Goodman, and Kittay (among others) who count the creators of metaphors as often straightforwardly aiming to communicate something appropriate to the words they use, and as responsible for the truth of what they assert. When Romeo says that Juliet is the sun, or when the critic says that Tolstoy is an infant, they are each using language to make assertions whose content makes essential use of our knowledge of the meanings of the expressions used, and which might be true, or might be false.19 This applies equally to high-octane ‘poetic’ metaphors, even though special factors might well intervene to make the mundane, give-and-take practices of assertion inappropriate. The point about truth is not that we always insist on it in metaphorical utterance, but that we must find something to apply it to in those cases where it does figure. In this regard, it should be noted that even unarguably literal claims can be made in poetic contexts without our bothering over their truth or falsity. A poet who writes about certain flowers blooming in May is not thought to be speaking as a botanist.

Given its occurrence in a play, Romeo's remark in fact does occur in a truth-irrelevant context, but, for the purposes of the example, I have been following the usual convention of treating the remark as if made by a non-fictional character keen on conveying a truth about the object of his affections. Standing below the balcony, Romeo could have said something non-metaphorical about Juliet and, by that convention, this would have been taken as an assertion. (Imagine that he said: ‘I love that woman.’) So, bracketing the theatrical context, there is no reason, intrinsic to (p.122) the metaphor, why we should regard (R) as anything other than an assertion for whose truth Romeo takes responsibility.

What makes Romeo's remark truth-apt is his qualifying use of the hybrid predicate ‘is the ↓sun↓’, and though this begins with a call on linguistic knowledge, knowledge of the copula, and the descriptive phrase ‘the sun’, the call on his and our knowledge goes beyond this. How far beyond? And in precisely what direction? These questions are in essence those I raised earlier about the processes of semantic descent and qualification, and, as noted, the plan is to deal with them later on. Still, I have to say something here about these questions, even if it is only to give the merest sketch, since I am aware that there might be a certain scepticism about the assertoric credentials of (R), understood in the way suggested.

The source of this scepticism is likely to be a certain model of linguistic understanding which, if left unchallenged, can make it seem as if there is a vast, even unbridgeable, difference between ordinary assertions and metaphorical assertions, at least as the latter figure in my account. Since it would take more than a chapter of another book to deal with this thoroughly, my devoting only part of a section to it in this one suggests rightly only the merest outline. But that outline is necessary, and I hope to make it clear enough for one to imagine how to fill it out.

Consider this perfectly ordinary assertion that Romeo might have made:

(17) Juliet is a woman.

What would we count as showing that someone understood an utterance of (17)? It is all too easy to think that in understanding what is said about Juliet here, what is required (at least in part) is a grasp of something—a meaning—which determines, among other things, how to sort items into those which fall under the concept woman, and those which do not. Or, since talk of meanings is out of favour, that what is required is that an interpreter bring to the context of utterance his or her knowledge of the contribution to truth conditions of the predicate expression ‘woman’, where this contribution is thought of as something which is in principle capable of dividing the world into those things which do, and those which do not, satisfy this predicate expression.

Of course, these can be perfectly innocuous claims: they may be taken as merely convoluted ways of insisting that anyone who understands (17) must know that it asserts of Juliet that she is a woman. But if one isn't careful they can be taken as the preface to something thoroughly misleading. I have in mind here the model of understanding which takes too seriously the idea of there being such a thing as the meaning of the predicate (perhaps its extension-determining power), and imagines it as a device which an appropriately trained speaker has somehow stored up in his mind, and is able to deploy when required. The trouble with this has nothing to do with the idea that such a device might be an abstract or mental object, and everything to do with the job of work assigned to it. For the idea of a meaning as a device which somehow contains the principle of sorting that goes with ‘woman’ is one that has been rightly criticized by Wittgenstein (and many others).

(p.123) Of course, it is controversial that there is a problem with this story; many still find it worth telling, even given the Wittgensteinian rule-following arguments. (As far as I can see, psychologists tell this sort of story all the time, not even noticing that the Wittgenstein who gave them family resemblance to play with only a few pages later offered trenchant criticism of a use that has come to be made of family resemblance in the project of representing concepts.) In any case, my aim here is not to detail the arguments against the view—as noted, that is not a task for this book—but simply to warn against allowing it to influence, perhaps subliminally, your understanding of my account of metaphor assertion. For if you think that the predicate in (17) can only be understood by someone who has got hold of some such thing as its meaning or sense, and if you think of this meaning as some sort of device for determining the application of the predicate, then you are apt to be particularly unhappy about my account of Romeo's metaphorical assertion. You are apt to point out that there is nothing in the use of the object—the sun—which corresponds to such a meaning, nothing which fixes a range of application of this object when it is embedded in the linguistic framework of predication. Moreover, in not finding anything that answers to the meaning of qualifying object, you are apt to question anyone's taking Romeo's (R), interpreted in my way, as an assertion.

Several examples will show what I am up against. First, imagine a loyal retainer to the Capulet household who overhears Romeo's utterance of (R) and says:

It is early and Juliet has just come out on the balcony. Romeo says that she is the sun, but he is deceived. Most mornings Juliet sleeps in until nearly noon: she is after all a teenager.

Clearly, it is extremely tempting to describe such a case this way: the retainer takes Romeo to be asserting a thought, but it is not the one we imagined him as expressing. Hence the retainer's assessment of falsity is irrelevant, and communication non-existent. If in this case the retainer had made a simple mistake—perhaps just mishearing Romeo's words—then there would be no problem. But the worry hanging over the semantic descent account of metaphor is the possibility that this kind of thing could be the norm in cases where qualification figures, and hence that the qualifying object simply does not properly fix the extension of the hybrid predicate. Here is a second example. In correspondence, Jerry Fodor (helpfully?) suggested this version of Romeo's (R):

Juliet is a real knockout; hot stuff,

and, while this might somehow seem more on track than ‘early riser’, it too is unlikely as a rendition of what Romeo said.20

What lies behind these examples is something like this. In the ordinary case of (17), we can describe an interpreter as having worked out that:

(18) Romeo said that Juliet is a woman,

(p.124) because this credits the interpreter with the possession of, among other things, the meaning of ‘woman’, and we can rely on anyone who, as it were, owns this meaning to be able correctly to sort things, in virtue of this meaning, into those which satisfy the predicate and those which do not. However, when it comes to (R), my recommendation is that we describe an interpreter as having worked out that:

(19) Romeo said that Juliet is the ↓sun↓,

but this doesn't seem to come with the same guarantees as (18). That is, (19) does not credit the interpreter with possessing a meaning associated with the sun which would rule out the retainer's and Fodor's unacceptable renditions. Indeed, the worry is that (19), for all that I have said so far, could be used to characterize both the retainer's and Fodor's versions, even though these are plainly enough conflicting.

Responding to this, I suggest that the appeal-to-meanings gloss of the straightforward (18), the sentence we use to report an interpretation of (17), is misleading, and this for broadly Wittgenstein's reasons. However, when we understand it correctly, the way is also open to seeing (19) in a more flattering light.

When (18) is asserted by some interpreter, what it reports is not the possession of some extension-determining element called the sense or meaning of the predicate ‘woman’. Rather, it reports the fact that, by our lights, the interpreter has made sense, in a specific way, of the speaker's action in producing just those sounds—that the interpreter has managed to fit those sounds into a larger network of attitudes and actions. Here is a way of spelling this out:

The adequacy of the total theory [of sense or meaning] would turn on its acceptably imposing descriptions, reporting behaviour as performance of speech acts of specified kinds with specified contents, on a range of potential actions—those that would constitute speech in the language—describable, antecedently, only as so much patterned emission of noise. For that systematic imposing of descriptions to be acceptable, it would have to be the case that speakers' performances of the actions thus ascribed to them were, for the most part, intelligible under those descriptions, in the light of propositional attitudes; their possession of which, in turn, would have to be intelligible, in the light of their behaviour—including, of course, their linguistic behaviour—and their environment. The point of the notion of sense—what the content-specifying component of a total theory of that sort would be a theory of—is thus tied to our interest in understanding—fathoming—people. We have not properly made sense of forms of words in a language if we have not, thereby, got some way towards making sense of its speakers. If there is a pun here, it is an illuminating one. (McDowell 1998: 172)

Going perhaps a bit further than this suggestion, but I believe in the same direction, I would describe (18) as a marker of a kind of co-ordination that exists within a linguistic group. The group consists of the speaker of (17), the interpreter of (17) who produces (18), and we who underwrite the interpreter's (18) as a correct interpretation of (17). In effect, our preparedness to accept (18) as true indicates our confidence that Romeo's action in uttering (17) does in fact fit intelligibly within the overall complex of his, the interpreter's, and our own actions and thoughts. We can say that the acceptability of (18) shows there to be a kind of attunement amongst all the participants. Note though that it is one thing to consider (18) as announcing or (p.125) marking such attunement, and another to insist that it, or elements in it, bring that attunement about. However tempting it is, (18) should not be thought of as introducing an element, for example, the sense of ‘woman’, whose possession somehow dictates co-ordination amongst the relevant parties. Though there is nothing about Wittgensteinian exegesis that is beyond dispute, I think that this is not only an important element in his rule-following considerations, but is one that could and should be taken on board by philosophers of language. In any case, it is something that I shall accept here, leaving further arguments in its favour for another time and place.

If we look at (19), that is:

(19) Romeo said that Juliet is the ↓sun↓,

in a parallel way, the problem raised by the retainer's and Fodor's versions turns out to be more apparent than real. For if we take (19) as announcing attunement amongst Romeo, an interpreter, and ourselves, rather than as having the role of imposing such attunement, nothing prevents our insisting that, as far as the retainer and Fodor are concerned, (19) is just not warranted. The logic of the original objection went like this: the retainer and Fodor were clearly mistaken about Romeo, but, since each could be characterized by (19), this attribution is just too thin to be an account of what Romeo said. To which I reply: given the proper way to understand such attributions, and the fact that we understand straight-off just how mistaken both the retainer and Fodor are about Romeo's (R), nothing compels us to employ (19); if there is no attunement, then there is no grounds for asserting what is, after all, by our lights a marker of attunement.

I can imagine someone thinking that this reply shows a certain perversity. Surely, the idea behind the objection is that the retainer and Fodor can both sign up to (19), because they both do in fact recognize that what is being said is that the sun qualifies Juliet. They do see that what is involved is a metaphor, and they cannot be accused of merely taking the original (R) to be a kind of literal, coded way of saying either that Juliet gets up early or is hot stuff. But, having signed up to (19), each of them goes on to make comments about Romeo's assertion that show them to have misunderstood. And the fault lies, so to speak, not with them but with my account of what Romeo said.

There are two things that can be said in reply here. First, I think that this way of putting the objection presumes just the kind of demand on interpretative attributions that I laboured to discredit. (It also shows how easy it is to slip into the mistake of finding such a demand reasonable.) It is simply not the case that when we find an interpreter able to assess a certain utterance as an assertion with the content, say, that a is an F, we are thereby crediting that interpreter with possessing a device which itself correctly determines the application of F. The right picture is really quite the reverse. It is because, by our lights and against a background of intelligibility-conferring attributions, the interpreter has got hold of a certain way of treating things as F, and shares this with the original speaker, that we find correct the interpreter's assessment of the speaker's utterance. Given what is in any case presumed by the objector (p.126) to my account, that both the retainer and Fodor do not share an understanding of (R) with Romeo, we should have no hesitation in resisting using our (19) in characterizing their understanding of his assertion. Insisting that the retainer and Fodor both know that (R) is a metaphor and even perhaps that it involves the use of a hybrid predicate (‘is the↓sun↓’) to qualify Juliet, is not enough to make (19) appropriate. It would be enough if the hybrid predicate was a device which itself fixed this application, so that anyone who, as it were, owned it—who knew that it was operative—couldn't make the retainer's or Fodor's mistakes. But this is precisely the picture that we should learn to ignore.

The second point to make in reply is somewhat more concessive. I do realize the worry about my account of Romeo's assertion turns in part on a difference between ordinary linguistic predicates and hybrid or proto-predicates. With the ordinary ones, for all that they keep philosophers of language up at night, we use them pretty unreflectively in making assessments of understanding. Romeo utters the sentence:

(17) Juliet is a woman,

someone hears this and says:

(18) Romeo said that Juliet is a woman,

and we have little hesitation in thinking that the interpreter got it right. The connections between, for example, Romeo's and the interpreter's linguistic actions using the word ‘woman’, and the no less important connections to the myriad further actions and attitudes of all those who use ‘woman’, are simply and succinctly crystalized in the deployment of the word itself. In contrast, there is no such easy route to the characterizaton of Romeo's claim in (R). However, even while conceding this, I insist that this difference is no more than one of degree.

Think of how it can go wrong, even in the purely linguistic case. I have in mind here things the interpreter might go on to say which could make us wonder whether he and Romeo should be seen as occupying the same place in, as it is useful to put it, the space of reasons that the use of ‘woman’ marks. (Using Sellars's metaphor-laden terminology helps to reinforce the earlier point about attributions: Romeo and the interpreter do not come to occupy a location in the space of reasons fixed for them by their use of certain words. Rather, the words have the significance they do in virtue of speaker's and hearer's occupation of some such place.21) For example, if the interpreter of the ordinary (17) went on to claim that Romeo was thereby saying:

(20) Juliet is wilful,

or:

(21) Juliet is subservient,

(p.127) we might begin to wonder whether, as it is often put, he had the same concept of woman as Romeo. Of course, we can wonder this without having yet to give up on (18) as a correct interpretation. However, this is only the beginning, and there are other ways of going on that might leave us less sure about the matter. Suppose that this same interpreter insisted:

(22) Romeo said that Juliet was the offspring of his uncle.

This might make us wonder whether, from our point of view, the retainer had somehow confounded ‘woman’ and ‘cousin’. And more bizarre continuations can be imagined—continuations which would make it impossible to maintain the pretence that ‘woman’, as it figures in (18), is the right word to use in characterizing the interpreter's report of Romeo's utterance.

With my treatment of Romeo's (R), these same kinds of problem are perhaps more easily conjured up; suggestions like those of the retainer and Fodor do not require much imagination. But what is important is that they present problems no different in kind from those we can come up with by applying our imagination to the linguistic case.

On my account of (R), there is no way to capture the content of Romeo's assertion by using purely linguistic predicates. The linguistic means we employ to succinctly characterize the place in the space of reasons which, for example, Romeo in uttering (17), and his interpreter in uttering (18), occupy, just don't work for (R). Nonetheless, this should not lead us to abandon the hybrid-predicate characterization of that content. The predicate expression ‘woman’ contributes to the content of (17) because it is treated as reserving a particular place in the space of reasons; it functions, or is rather allowed to function, without the need for further commentary.22 The hybrid ‘is the ↓sun↓’ does not get this same treatment—that was conceded above—but, when we look a little deeper, we can see it as no less fit for purpose. The very fact that we can see straight off that the retainer's and Fodor's suggestions are wrongheaded is evidence of this.

There are things we know (or believe) about the sun, as well as about the extra-linguistic and linguistic context in which Romeo's assertion was made, which should be seen as background to our use of the hybrid predicate in (19). These are precisely the sorts of thing that I will say more about in the last section of this chapter, but it is important here to have the right view of them. I am not saying that the things we know about the sun etc. are a substitute for things said to be known by language users in respect of ordinary predicate expressions like ‘woman’. Having resisted the picture of understanding ordinary predicates as the possession of devices (concepts?) which impose their correct uses on speakers, I am scarcely looking to fill in (p.128) a similar picture in respect of hybrid predicates. The point about those things we know about the sun, and about the background to Romeo's utterance, is not that they constitute such device, but that they help us identify the place in the space of reasons appropriate to the hybrid predicate as used by, and of, Romeo when he utters (R).23 For all that we put our faith in the use of ‘woman’ in certifying the attribution in (18), that faith is grounded on the background knowledge we have of the things people think about the world, about each other, and of course it is no less dependent on the things they do, and the reasons for which they do them. Similarly, our faith in the hybrid predicate's potential to reflect the content of Romeo's (R) is grounded in such knowledge, notwithstanding a tendency to think that purely linguistic predicates can do their interpretative work without explicitly calling on any such outside help. Indeed, it would be surprising if there were not some such difference between literal assertions like (17) and metaphors. For if there weren't such a difference, metaphor would simply not be the phenomenon it is. (I shall say more about this in the next section, where I consider the issue of paraphrase.)

More promissory notes have been issued, but the upshot of the section has been a defence of our using down-arrow representations in specifying the truth-relevant content of the assertions made with metaphors. These representations reflect the background against which the qualifying use of the appropriate object is made. (p.129) They do not by themselves impose uses of the necessarily hybrid predicate that figures in the representation of the metaphor. But that is not something we should expect, even of purely linguistic predicates as used in specifying the content of common-or-garden literal assertions. Whether we are speaking of purely linguistic predicates or my hybrid ones, these figure, not as ways of imposing certain patterns of use, but as marking an attunement in thought and action between speaker, hearer, and all other participants in the space of reasons.

3.7. Paraphrase

According to my account, understanding a metaphor consists in fastening onto an object-exemplar of a relevant predicate, and then taking that object as a qualifier of the metaphor's subject.24 In this way, a hearer who is attuned to the predicative use of that object comes to understand something about the subject of the metaphor, while justifiably considering the speaker responsible for the truth of that information. In sum, all the ingredients of common-or-garden assertion are present in metaphor. However, since the information is conveyed using a hybrid predicate—one which contains an object in a qualifying role—it would be bizarre to ask someone to express this same information in other words. Admittedly, it is words in the metaphor that call on the object. But it is what I have also called the ‘proto-predicate’, object included, which conveys a message, not the words themselves. Since the speaker is using an object, not words, to convey a message, it makes no sense even to try to paraphrase a metaphor in the strict sense of the term.

A useful way to think of it is that, on the semantic descent account, a metaphor functions like a picture, diagram, or map. As we saw in Chapter 1, a request to paraphrase a picture makes no sense, but then again neither does a request to paraphrase an object. Nonetheless, there is plenty of house-room in the semantic descent account for the other activities with which paraphrase is all too easily confused. I am referring here to translating, elucidating, and generally commenting on metaphors.

3.7.1. Paraphrase and translation

As already noted, it is not usually all that difficult to translate a metaphor into another language. This somewhat mysterious fact has been noted by many writers, but has never been to my mind satisfactorily explained. While native English speakers find it hard to imagine how, in general, the power and beauty of Shakespeare's language can survive translation, we do not have the same difficulty with certain bits of that language, bits such as Romeo's description of Juliet. Whether she is ‘le soleil’, ‘il sole’, or ‘die Sonne’ makes little difference to the impact of the original metaphor. Nor, on the semantic descent account, would we expect it to. For, on that account, what is being translated is the word or phrase (p.130) before, as it were, descent to the proto-predicate takes place. So long as these translations preserve reference we should expect, rather than be surprised by, their adequacy.

One must be careful here: a superficial reading of my claim about preservation of reference can make it seem vulnerable to obvious counterexamples. For example, assuming that the sun is my favourite heavenly body, does this mean that we can translate Romeo's assertion into French as:

(R1) Juliet est le corps céleste favori de SG?

Or, given that the sun undergoes nuclear fusion, does this make the following a good translation:

(R2) Juliet est un four nucléaire, autour de qui la terre fait sa revolution?

Clearly, the answer to both of these questions is ‘no’, but then neither is really a serious counterexample to my original claim. What I contend is that, in cases where the words in a metaphor in a source language are replaced by their standard or usual translational counterparts in a target language, the semantic descent account can explain the surprising fact that metaphorical effect is preserved. The standard translational counterpart of ‘the sun’ in French is neither ‘le corps céleste favori de SG’, nor ‘un four nucléaire, autour de qui la terre fait sa revolution’. So, though reference is preserved when these expressions are substituted for ‘the sun’, they are not reference-preserving translations.

Couldn't there be cases in which the only available way to render ‘the sun’ preserved reference, but led to hopeless translations of the metaphor? Suppose, for example, that in some language, Native, the best you could do for ‘sun’ would be something which came back into English as: ‘the evil staring eye of the Ox-god who rises from bed every morning’. (For whatever reason, the speakers in this community regard the sun as threatening and malevolent.) Clearly, this:

(R3) Juliet is the evil staring eye of the Ox-god who rises from bed every morning.

disfigures Romeo's remark. But doesn't it also undermine my claim about translation? Someone might insist that we have in (R3) a translation which preserves reference but doesn't cope with Romeo's metaphor.

Though the matter is somewhat intricate, I think this possibility is no real objection to what I have claimed. My defence proceeds on two fronts. The first of these is straightforward: I have not insisted that every putative translation of an expression preserves the metaphorical content of the original; merely that when there is preservation, it can be most naturally explained by the semantic descent account. The second defensive front is less straightforward: there is good reason to wonder whether we should accept that, in respect of ‘sun’, translation between English and Native is genuinely possible. And, even if we do accept that translation is possible, the result is more supportive than undermining of semantic descent. More detail on both strategies follows.

(p.131) When a hack, English–French translator, for example, one who makes a living translating washing-machine manuals, gets to work on a text of Shakespeare's, we would expect the result to be less than lovely. Yet, even with such low expectations, we would find that some bits of the text, for example, Romeo's metaphor, pass through the translation process with little or no loss. This is surprising enough to call for comment, and what I have been claiming is that the semantic descent account can explain, without fuss, why this happens when it does. I should not be taken as having made an open-ended claim about the possibility of translating every metaphor into any language. So much then for the first line of defence.

Is it appropriate to regard (R3) as a translation? The example was under-described, so it needs filling out. One possibility is that Native speakers have no word for the sun, always using a phrase in their language which, when retranslated back into English, leads us to (R3). In this case, I think we would very much doubt that English–Native translation of ‘sun’ is possible. The need for a circumlocution in one language replacing a single word in another is usually a good indicator of untranslatability, and if (R3) is not a translation in any reasonable sense, then there is no case for my account to answer. (That this could happen with ‘sun’ is highly unlikely, but there is nothing incoherent in making the assumption that it could happen.)

The other possibility is that, in addition to the ‘Ox-god’ descriptive phrase, there is a single word for the sun in Native, and hence that translation would seem possible. Strictly speaking, this would mean that something had gone wrong with the retranslation of (R) back into English as (R3): if there were a single word in Native for sun, the retranslation should come back as just (R). Yet, as the case was described, it doesn't seem quite acceptable to say nothing more than that retranslation gives us back (R). Clearly, the case deserves a closer look.

From the perspective of the semantic descent account, the one-word translation of (R) into Native lets us down, not because anything has gone wrong linguistically, but because the attitudes of Native speakers towards the sun, the object itself, differ so strikingly from our own. This is shown by their readiness to employ a phrase that comes back to us as in (R3). That this can happen is familiar enough: there are non-hypothetical cases in which translation is unhelpful, in spite of referential overlap in vocabulary. French and English have words for cabbage, but the French use this vegetable to characterize human beings in a complimentary way, while this is most certainly not something done by English speakers. If something like the vegetable scenario were true of Native and English in respect of the sun, then it would certainly be unreasonable to deny that translation is possible. But it would be no less unreasonable if we thought of these translations as preserving metaphor.

Does admitting this undermine the idea that semantic descent explains the relative ease of translation of metaphor? The answer is ‘no’, partly because of my first line of defence: semantic descent explains why it is often easy to translate metaphors, but this does not imply that translation always works, let alone that it is always easy. But there is more to it. For not only does the semantic descent account explain translation when it is easy, it also gives what seems the right explanation for (p.132) failure in cases where translation doesn't work. And this is something that many other accounts get wrong.

Black, and many Content Sufficient theorists who follow him, try to account for metaphors like (R) by complicating the story we tell about the meanings of the words. Very roughly, Black invites us to think that words not only have minimal and narrowly fixed (literal) truth conditions, but that language users are able to construct further meanings for these words on a case-by-case basis as these words occur in metaphors. The crucial thing to note is that the focus of all this extra effort is words: it is the familiar ‘sun’ that comes to have an unfamiliar meaning in (R). Thus, when someone attempts to translate (R)—which, from our point of view contains ‘sun’ with its newly minted meaning—into Native and then back into English, that result is bound to be unsatisfactory, given what the natives think of the sun. For this reason, a Black-style theorist will come to think that translation isn't possible, even if Native has a single word for the sun. There is thus no room in a Black-style account for the obvious point made earlier that, if Native has a word for the sun, the problem is not translation of ‘sun’, but rather the attitudes that Native speakers have towards the object that ‘sun’ picks out. The semantic descent account makes objects the focus of metaphor, whereas for most other Content Sufficient accounts this focus is words, a difference that is actually quite important. (Something like this same issue will figure in Chapter 5 when I discuss Josef Stern's account of metaphor.)

3.7.2. Paraphrase and other near relations

Moving on from translation, I turn now to something merely touched on in the treatment of metaphorical truth. Metaphors—especially but not only of the literary variety—tend to provoke explanatory commentaries, and it is crucial to understand how these commentaries relate to what I have been calling ‘paraphrase in the strict sense’. For unless I clarify this, there will always be the suspicion that I am perversely hanging onto a much too strict idea of what is involved in paraphrase.

Here are some of the things that might be produced in respect of Romeo's remark:

(23) i. Romeo thinks that Juliet is necessary to his very existence.

ii. Romeo thinks that Juliet is responsible for his seeing the world aright.

iii. Romeo thinks Juliet is time itself.

(24) i. The sun is the ultimate source of light and warmth.

ii. The sun is the measure of time.

iii. The sun makes life on earth possible.

Though I have perhaps been overzealous in displaying these comments as belonging to sharply bifurcated classes, there can be little doubt that some such division exists. Some comments sound like glosses on what a speaker/writer aimed to achieve in using the metaphor, whereas the others take aim at the metaphor itself. Neither should be seen as a way of paraphrasing what the speaker said, though it is often difficult to resist this when the comments take the form of (23i–iii). My suggestion (p.133) is that we can make best sense of (23)–(24), and their difference from paraphrase, from the perspective of semantic descent.

As I have insisted, everything begins with our taking seriously the idea that Romeo made an assertion, that he expressed a thought he regards as true. On the semantic descent account, this expression of thought, achieved as it is through metaphor, is special: the thought expressed is available to a hearer who is attuned to the metaphor, but we should not think that this requires a hearer to be able to find other words that express that same thought. This is because Romeo's thought is conveyed by a proto-predicate—an object pressed into service as a qualifier of Juliet. Quite simply: though Romeo used familiar words in expressing his thought, the thought itself was not expressed by those familiar words.

That said, Romeo's assertion is undeniably an intentional action, something undertaken for a reason. It can be seen as aiming both at expressing something about Juliet and (in principle) at conveying that something to an audience. (One must put on one side here the fact that Romeo is, as it were, talking to himself. The complexities wrought by the fact that this particular assertion occurs as part of a monologue in a play are interesting but not relevant to the general point being made with this example.) Now, as with any other action, it always makes sense to wonder why it was done—to ask after the state of mind of its agent. Moreover, given that we are here dealing with a linguistic act which, unlike, say, making yourself a cup of tea, is expressive and informative, it makes sense to wonder about the effects that it can have on an audience.25

Putting these considerations together, it is unsurprising that commentaries take the form they do. The examples in (23) are typical of the attempts one might make to understand the state of mind which led Romeo to say what he did, whereas the examples in (24) suggest things about the metaphor object which might be the basis of attunement between speaker and audience. However, in neither case should we see these claims as aiming to capture the thought that Romeo expressed; in neither case is the aim to put Romeo's assertion into other words. Still, it is easy to mistake these attempts for paraphrase, especially when it comes to the items listed in (23). Such a mistake might take the form of someone's insisting that I have left out an important category of ‘commentary’. The objector continues: in addition to (23) and (24), you should have included:

(25) i. Romeo said that Juliet is necessary to his very existence.

ii. Romeo said that Juliet is responsible for his seeing the world aright.

iii. Romeo said that Juliet is time itself.

This is because at least part of the reason Romeo has for uttering the sentence: ‘Juliet is the sun’, is that he thinks that Juliet is necessary to his very existence. So, we should count this thought as part of what he actually expresses when he does (p.134) produce the metaphor. By doing this, we can achieve paraphrase, albeit partial and approximate, by uttering the sentences in (25).

My first reaction to this—one that I expect is not idiosyncratic—is that there is something distinctly odd about the sentences in (25). It is certainly right to think that Romeo asserted something, but it doesn't seem right to think that he actually said any of these things. There are of course complex (and disputed) constraints on accurately reporting speech, but they are surely not met by any of (25). To sharpen this perception, suppose you hadn't heard Romeo's remark, but were told:

He said that Juliet was necessary to his existence.

What range of words would you then imagine Romeo to have used? Not the ones he actually used, I suspect. Of course, my finding (25) odd might well be an artefact of my advocacy of the semantic descent account. So, it is important to say something more.

The sentences in (23) are a sample of the kind of thing that Romeo might well have thought, and knowing these thoughts certainly helps us understand why he said what he did in uttering (R). However, there is no reason to think that thoughts which in this way explain an assertion are themselves expressed in it. To be sure, this mistake is easily made: a reason for my saying: ‘The meeting was short but tedious’ is that I thought it was. Here the thought behind my utterance, and the thought expressed in it, are one and the same. But, as should be obvious enough, this is by no means generally true. For example, thoughts that explain an assertion might in some way imply, or be implied by, though not be the same as the thought expressed in it. Or, they might be thoughts about some end, the envisaged achieving of which brings about the expression of some other thought, perhaps one characterizing the means. My suggestion is that metaphor is more plausibly counted as in the class of assertions whose supporting reasons are distinct from the thought expressed. Moreover, this suggestion is not dependent on acceptance of my account. Everyone agrees that the thoughts expressed in, or by, the words of a metaphor are in some way distinct from thoughts expressible without metaphor. Indeed, the perennial mystery and appeal of metaphor is in large part traceable to this elusiveness. If for no other reason we should therefore be ready to find sentences such as those in (25) off-beam. That Juliet is necessary to Romeo's existence may well be a thought that helps us gain a deeper insight into Romeo's remark and/or one which is part of his reason for making it, but this wholly non-metaphorical claim can scarcely be a whole or even a part of what Romeo actually asserted.26

The temptation to think that perfectly apt commentaries on metaphor are also ways of paraphrasing them is very difficult to resist. After all, the aim of commentary (p.135) is a deepening of understanding of both a metaphor and its author. Without care, it would be easy to slide from ‘understanding’ in this broad sense to ‘understanding’ as, more narrowly, grasping what is said. The upshot of this would be that commentary becomes paraphrase. But if one holds fast to the idea that a metaphor is in some way a device for communicating thoughts distinct from any purely word-based one, then there is plenty of scope for allowing commentary that is in no way paraphrase.

Just so that we can keep track, I shall call comments like those in (23) rationalizations of a metaphor, and those in (24) elucidations. I do not insist that this classification is exhaustive, though it is certainly generous enough to allow in most of the kinds of thing you are likely to come across in discussions of any specific metaphor. In the present context, the thing to note is the way in which the semantic descent account not only shows us how to keep these two kinds of activity separate from paraphrase, but also how to make the distinction between rationalizations and elucidations.

Rationalizations and elucidations can take various forms and, while this is not the place to sift through actual commentaries, sorting their contents into these two categories, we can count the sentences in the somewhat artificial lists (23) and (24) as pointing out the way to do it. What is suggested by these two lists is that: (i) rationalizations are essentially non-metaphorical attributions of thoughts about the subject of the metaphor, either to the speaker (as shown in 23),or to an audience (not shown, but easily imagined); whereas (ii) elucidations focus, not so much on this traffic of thought, but on features of the metaphor that make such traffic possible.27

While one can easily imagine commentaries that combine these two features, they are nevertheless distinct, and the semantic descent account provides a straightforward explanation of this. Given the idea of metaphor as a device for reaching down through language into the world, and using what is found there to express and convey information, we should actually expect to find commentary taking (at least) these two forms. On the one hand, there would be rationalizations—considerations focusing primarily on what was called above the ‘traffic of thought’. Here what is in question is not so much the content of the metaphor assertion as the trail of thoughts that leads to and from it. On the other hand, there would be elucidations, comments which take aim at the metaphor itself and, as I see it, highlight the roles of the relevant objects in what are proto-predications.

Unlike the case of ordinary predication, in proto-predication one cannot rely on attunement marked by what are presumed to be shared words. In the case of (p.136) metaphor, attunement must be grounded on a shared appreciation of the significance of objects. Our attunement to Romeo's remark comes not from what we know about the meanings of words—though words and their context do play an important part in the process—but from what we know about the sun. In so far as this attunement is less than perfect, or perhaps simply to make it more explicit, one would find it natural to make the kind of comments represented in (24). Indeed, it is difficult to make sense of these sorts of comment except on the semantic descent account. (My comments on attunement have so far been rather teasingly vague; a more focused discussion of this will round out this chapter.)

Summing up this section: it has been argued that the demand for paraphrase in respect of metaphor is simply inappropriate. Admittedly, this is because I have been insisting on taking ‘paraphrase’ in a strict sense, but I have offered grounds for this. Coming at the matter indirectly, I have argued that the only real pressure for loosening our standards of paraphrase comes from two interrelated sources: (i) an unreasonably narrow model of assertoric communication, one which requires every thought expressed in such communication to be expressible, even if approximately, by words in natural language; and (ii) a natural, and therefore intelligible, but ultimately wrongheaded tendency to take the kinds of thing we say about specific metaphors as themselves paraphrase (though in a ‘loose’ sense).28 Taking these sources into account, and treating them with the suspicion they deserve, we can relieve the pressure on the notion of paraphrase. Neither rationalization nor elucidation of a metaphor is any more a way of paraphrasing than is translation. Moreover—and this is the most important bit—the semantic descent account helps us understand what is going on in respect of all of these notions. It suggests why translation of metaphor succeeds, against the odds for translation generally; it grounds what is a real distinction between rationalization and elucidation within the category of metaphor commentary; and finally it offers a straightforward reason to stick with the notion of paraphrase as it is ordinarily and strictly understood.

3.8. Transparency

I begin with a brief résumé of my discussion of this feature of metaphor, as it is the least familiar of the three, and hence most subject to misunderstanding. Also, some of the points made above in connection with truth and paraphrase now put us in a better position to understand transparency.

When you hear the utterance of a sentence in your native language, then, assuming nothing untoward about background noise or trouble with your ears, you (p.137) just do hear the meaning of the utterance. This is not something voluntary or optional: it is just not possible to suppress the meaning of what is heard.

This familiar fact is the beginning of research in psycholinguistics—research aimed at explaining why this is so and how it is accomplished—but the fact itself is something familiar, and not the upshot of such research.29 ‘Transparency’ is the label that I use for this feature, albeit with some misgivings, and what I insisted in Chapter 1 was both that transparency is as true of metaphor as it is of unarguably literal sentences and that many accounts of metaphor could not cope with this parity.

In showing why certain accounts of metaphor cannot cope with transparency, the claim itself needed some refinement. This did not involve resiling from my insistence on transparency as the involuntary hearing of meaning, but it proved necessary to exercise care in explaining the import of the phrase ‘hearing meaning’, in order to avoid either trivialization or implausibility.

There are two threats on the trivialization front. On the one hand, it is not enough to say that the meaning we take in on hearing appropriate utterances is simply literal. It would be convenient if this literality claim were true. When Romeo utters:

Juliet is a woman,

English speakers cannot avoid hearing this as having the content that Juliet is a woman. If in the same way, when Romeo utters the familiar:

(R) Juliet is the sun,

the content taken in is that Juliet is the sun, that is, that she is the very large far away object in the centre of the solar system, we would be home and dry as far as transparency was concerned. Admittedly, what would be heard as a meaning in this case would be something absurdly false, but it would at least be a meaning or content. And this might be felt to be enough to explain the parity in respect of transparency between the literal and the metaphorical.

Unfortunately, this just doesn't work. Aside from the fact that there is a real issue as to whether the literal reading of (R) is nonsensical, not simply false—whether, that is, there is such a reading—there is an even larger problem. While I do not want to claim that, in hearing the meaning of (R), an ordinary auditor gets right to the bottom of this utterance, it certainly does seem true that more is got out of it than the merely literal. This is shown by the fact that whatever it is we take in on hearing (R) is enough to make intelligible the kinds of things said by way of what were (p.138) described in the previous section as elucidation and rationalization. I won't here repeat the arguments offered Chapter 1 to support this claim, but their upshot was that there must be some, perhaps shallow, kind of genuine understanding which preserves parity in respect of transparency between the literal and the metaphorical.

What this shallow understanding comes to is not easy to describe, and this brings me to the ‘other hand’ in respect of the trivialization danger. For, as we also saw in Chapter 1, it is tempting to try to characterize this understanding by saying that auditors hear the meaning of the familiar words in (R), and also understand that they are to take (at least some of) these words ‘metaphorically’. But this is no improvement on the previous attempt. The problem is that, in finding there to be further work that auditors must do to get at the content—even the shallow content—of (R), transparency is lost. (Again, more detail on this was given in Chapter 1.)

Against the background of the failure of these simplest proposals for dealing with transparency—proposals that would make it trivially easy to satisfy—one has to be careful not to credit hearers with access to a content so profound as to be implausible. As has been mentioned, meaning is non-optionally heard in the utterance of (R), but it is most certainly not the kind of content that makes elucidatory or rationalizing commentary on the metaphor otiose. What transparency requires of accounts of metaphor is that they come up with a notion of content relevant to the metaphorical aspirations of utterances, but which nonetheless leaves plenty of further room for commentary on the nature of these aspirations.

It was at this point in Chapter 1 that I worked up a proposal based on Davidsonian materials to show that the transparency challenge could be met, a proposal I called the ‘Image Account’ (IA). Aside from its intrinsic interest, the IA has a special role to play in the present context, for I shall argue that there are overlaps in respect of transparency between the IA and my own account of metaphor.

3.8.1. Transparency: the image account and semantic descent

According to the IA, Romeo's (R) is not heard as some absurdly false literal claim, nor as a merely skeletal arrangement of familiar words accompanied by general advice to seek a metaphorical interpretation. In fact, it is not heard as a claim at all. Instead, we hear Romeo's utterance as:

(R′) Juliet as the sun.

What (R′) does is to put in front of us a way of thinking about Juliet, something we can call an ‘image’, so long as we do not focus solely on the idea of visual images. There are two features of (R′) which make it ideal for meeting the transparency challenge. First, (R′) suggests a depth of understanding that is just right for transparency. The image of Juliet as the sun connects to the kinds of elucidation and rationalization we are likely to make about this metaphor, but doesn't, implausibly, pre-empt them. Indeed, the IA allows for the possibility that someone could ‘get’ (R′), but be stuck for anything else sensible to say about it. Second, there are good reasons to think that someone might actually hear (R) as (R′).

(p.139) As noted in Chapter 1, when we hear (meaningfully) a literal sentence such as:

Henry is English,

it is not implausible to think that there is a something like a simultaneous activation of an image (again, not by any means a visual one) of Henry as English—an activation which, depending on the context, may be reinforced or extinguished. There is, after all, only a small lexical difference between the two sentences, and their contents resonate with each other. If you are told that Henry is English, this conjures up the image of him as English, and if you are offered only the image of him as English, it is difficult not to wonder whether he is English, even if this information hasn't been explicitly provided.

Of course, for all the talk of simultaneous ‘activation’, I do not intend to prejudge results of as yet unperformed psycholinguistic experiments. But there are experiments suggesting that we do in fact simultaneously ‘access’ multiple meanings of ambiguous words, and only later use context to disambiguate them. So, it is not that far-fetched to think we might do something similar with ‘Henry is English’.

As noted above, my original purpose in constructing the IA was to show that transparency could be accommodated, not to offer the IA as genuine rival to my own account of metaphor. After all, the IA fails to negotiate the truth aptness of many metaphors, and, as described above, it is limited to metaphors of subject-predicate syntactical form. However, the usefulness of the IA is not limited to showing that transparency can be accommodated. In accomplishing this feat, it happens also to provide materials that the semantic descent account can appropriate.

The IA sees the words of a metaphor as, so to speak, projecting an image of one thing as another, and inviting a comparison between the properties of the objects that figure in that image. In contrast, the semantic descent account sees these words as forcing us to confront an object—an object which then functions as a proto-predicate (i.e. qualificationally), conveying information to an attuned hearer. This difference aside, however, the IA and semantic descent share a crucial feature: they require for their respective projection and confrontation an initial grasp of the ordinary meanings of the words used, and both take this grasp to be all that is required. It is this that allows them, in their otherwise different ways, to accommodate transparency. Relying as they do only on the ordinary meanings of the words in a metaphor makes it plausible that there is something available right from the start which can confer intelligibility, and this is all that is needed for transparency.

That said, the differences between the accounts do matter. The IA requires a shift from the copula ‘is’ to the preposition ‘as’. This move invites hearers to explore, rather than to agree or disagree with, what is put before them. Of course, having made this shift, hearers can then use the words in the sentence with their perfectly ordinary senses: it is after all the properties of the objects picked out by these words that serve as the starting points of exploration. But, passing over the question of truth aptness, the shift itself, small as it can be made to seem, is more radical than that required by semantic descent. In the latter, the structure of the original sentence is preserved.

(p.140) While this way of looking at the matter seems to be to the advantage of the semantic descent account, one must be careful (and honest). As noted above, the shift required by the IA is the syntactic one involved in the move from the copula to the preposition; all the other words in the metaphor undergo no shift in meaning. In apparent contrast, while the semantic descent account does not require any global syntactical shift, and while it begins with the ordinary meanings of the words in the metaphor, it does end up envisaging what might be thought of as a meaning shift, not of the whole sentence but of certain of its words. Though it is the ordinary predicate expression ‘sun’ which figures in (R), it is the sun itself which does the work needed to make the metaphor intelligible. Does this apparently additional shift diminish the account's ability to handle transparency? Does the IA then come out better in this respect? As you would expect, the answers to these questions are negative.

Though my account lends itself to idea that, in (R), ‘sun’ is in effect ambiguous, the second sense of, for example, ‘sun’ in (R) is so close to the surface that understanding, at least in some immediate sense, should be effortless.30 It is surely plausible that on hearing ‘sun’, we have unproblematic access to the object for which this word stands. (More on this follows below in 3.8.2.) And given that the copula in the original sentences invites us to take this object as a monadic proto-predicate, we have all we need for an initial understanding of the metaphor, one that satisfies the requirement of transparency.

The IA accommodates transparency by taking the words in a metaphor with their perfectly ordinary senses, not to express a thought, but to put an image before us. Once it is understood that some such image is involved in ‘getting’ the metaphor, a hearer has all that is necessary to engage in further elucidation and other kinds of comment. For all that the semantic descent account differs from the IA, it accommodates transparency in much the same way. Words are taken to have their ordinary senses, but rather than images, these words lead us to objects. ‘Getting’ a metaphor—hearing it as meaningful—then consists in recognizing that the objects fulfil a semantic function and, as with the IA, even this minimal understanding meshes smoothly with further elucidation and comment.31

(p.141) 3.8.2. Psycholinguistic speculations

As suggested above, the IA shift from ‘is’ to ‘as’ is one that might well be unreflectively accessed by a hearer. In saying this, I was mindful of psycholinguistic experiments which show that multiple interpretations can figure right from the start of sentence comprehension. Still, since the experimental results I had in mind concern word meaning, not the meaning of whole utterances, I was stretching a point. With the semantic descent account no such stretching is required, since all that need be in play are the two interpretations of the expression ‘sun’; one, the wholly linguistic predicate, the other, the object specified by that predicate, but serving itself proto-predicatively. Connected as they are, it would be strange if they didn't resonate with each other from the beginning of any sentence comprehension task. Nor is this all that can be said about the differences between the IA and semantic descent that might show up in psycholinguistic experimentation.

As I have said several times, the time-course of comprehension of metaphor is not strictly relevant to transparency, since transparency is about how metaphors strike us, not about how long they take to do so. Hence, my asides about these matters should be viewed as no more than speculative. Still, it should be counted as a mark against any account of metaphor that, in accounting for transparency, it forces one's hand about such matters before we even set out to do the experiments. Thus, it may be that metaphors on average take longer to comprehend than equivalently complex literal utterances, or it may be that there is no real difference in the time-course of the two tasks. Neither of these options should be closed off, yet such closing off seems inevitable on the IA.

The resonance of the two ‘meanings’ mandated by the IA—the literal and the image interpretations—is a resonance between contents of whole utterances. In order to understand Romeo as having presented an image of Juliet as the sun, we have first to confront the idea that she is the sun. That is, we have first to understand (R) in some sense other than that of image presentation. Given this, if the IA were correct, metaphor interpretation would be systematically slower than literal interpretation. This point needs careful handling.

Earlier I suggested that the image interpretation might become available during the comprehension of (R), and I still think this might be so. But even if this speculation were true, it would not eliminate the time-lag of metaphorical interpretation. Nor was that its point. Davidson, among many others, claims that we typically search for the metaphorical interpretation after we have found that the literal reading is unhelpfully false. If this were true, it would certainly require metaphor comprehension to lag systematically behind the literal, since the identification of metaphor would depend on the appreciation of falsehood, and the process of metaphor interpretation would then depend on that identification. It was to avoid the latter dependence that I outlined the possibility of a resonance between assertoric (literal) and image presentation contents associated with (R). With such resonance, there is no need for an auditor to hear the falsity of the literal as a starting gun for finding the metaphorical interpretation. Instead, factors having nothing to do with the falsity of literal assertion might reinforce the metaphorical interpretation (p.142) that accompanies the literal. But, though this might be a way to avoid the anyway implausible dependence of metaphor interpretation on recognition of literal falsehood, it does not deal fully with the time-lag issue.

Remember the psycholinguistically relevant issue is not the time taken over the literal and then the metaphorical comprehension of any given metaphor. Rather, it is the average time taken for comprehension of flatly literal sentences (of various syntactic complexities) in comparison with the time taken for sentences that are metaphorical (but match the literal ones in complexity). Clearly enough, if there is a dependence of metaphor comprehension on the discovery of the falsehood of the literal, then metaphors are always going to take, on average, longer to comprehend than merely literal sentences. But even without this dependence, the IA still requires some kind of whole-utterance interpretation, or even, merely, appreciation, as a step on the way to revealing the image that resonates with the literal, and this creates a built-in time-lag in comprehension.

The effect of this can be seen even in a literal sentence. Presented with ‘Henry is English’ we have to get some way to understanding it—though without necessarily evaluating its truth—in order to be in a position to switch our attention to its embedded, image-presentation form: ‘Henry as English’. It would be a speculation too far to say what effect this is likely to have on the actual time-course of comprehension, but the very fact that the whole sentence has to be processed in some way in order for the image to be appreciated makes it likely that the image interpretation would take longer to process. And if this is true of a literal sentence, it is bound also to be true of a metaphor. In contrast, the semantic descent account is not committed one way or the other to any difference in processing times.

On the one hand, it could be the case that we need to process the whole of Romeo's metaphor before we take on board its embedded proto-predicative use of the sun. It is even possible that, in spite of my scepticism, something like falsehood or anomaly are needed to trigger the move from the predicate ‘is the sun’ to ‘is the ↓sun↓’. In either case, we would then expect a systematic time-lag in the processing of metaphors as compared with literal sentences. On the other hand, the predicate expression ‘sun’ might well be taken in the way I suggested earlier, namely as having two related meanings that are both activated before whole-utterance comprehension. If this happens, then there would be no reason why metaphors would take systematically longer to comprehend. In this they would be like the two interpretations of ‘John found a bug in the room’, both of which are available, so to speak, at the same time, precisely because the activation of both senses of ‘bug’ takes place before any attempt is made to process the whole sentence.32

There is a final point necessary to round out this section. Because the IA requires the shift from ‘is’ to ‘as’, there is no obvious way to adapt it to the demands of metaphors not of the subject-predicate form. While it might be possible to cook (p.143) something up for more complex cases, it wouldn't have the elegant simplicity of the original IA—the changing of a single word—and that simplicity is crucial to its success with transparency. As far as the semantic descent account is concerned, the prospects are brighter. So long as the account can negotiate the move from subject-predicate to more complex forms—the first order of business in the next chapter—there should be no special problem with transparency. This is because all that the semantic descent accounts needs for transparency is the kind of bisemousness shown by ‘sun’. Whatever the syntactical position of the word or words that are the subject of descent, so long as there is the possibility of both readings being available from the start, transparency will be preserved.33

3.9. Attunement

A key ingredient in the semantic descent account is the idea that objects can function in very much the same way as linguistic predicates. An obvious challenge to this proposal is that this use of objects is not sufficiently constrained to fix the content of any assertion that might be made with them. Already touched on, this challenge bears re-statement, as a preface to facing up to it more directly than I did in section 3.6.

When someone says:

Juliet is a woman,

the content of this assertion is said to be fixed (at least as far as the predicate is concerned) by the meaning of ‘woman’. However, when Romeo says:

(R) Juliet is the sun,

then, in order so much as to countenance the idea that the sun can take on a predicative function, we must find something that fixes the contribution of the sun to this assertion, something which achieves what meaning does in the case of ‘woman’. Unfortunately, the challenge pessimistically concludes, it is difficult to see how any such thing could be provided. The sun is, as one might say, a mere object, not something which possesses the semantic properties necessary to fixing the content of assertions in which it figures. In respect of (R), this is shown graphically by possibility that there could be auditors who hear this utterance perfectly well, and even recognize the role that I claim for the sun, while still getting completely the wrong end of the stick about what Romeo is saying.

In section 3.6, I insisted that this pessimistic assessment takes too literally the idea that linguistic predicates come equipped with meanings, roughly devices the (p.144) mastery of which co-ordinate interpretations amongst speakers and hearers. Instead, I suggested that we should think of meanings, not as bringing about such co-ordination, but as reflecting the co-ordination that happens to exist. When an interpreter says:

Romeo said that Juliet is a woman,

and Romeo agrees with him, and we agree with both of them, all of this is simply a way of displaying the fact that, in so far as ‘woman’ is concerned, our linguistic actions (and the reasons for them) are co-ordinated with one another. In the language I used in section 3.6, the use of ‘woman’ in these interpretative claims marks out a place in the space of reasons occupied by all of the relevant parties. Or, putting it in a less fancy way, what this shows is something that is as close to common sense as one gets in philosophy of language: speakers and hearers in fact use ‘woman’ in the same way.

The upshot of these considerations is that we should not feel disappointed, even cheated, by the fact that objects don't also come equipped with meanings which impose order on those who use these objects predicatively. Someone hearing Romeo's (R) might very well get the wrong end of the stick about Romeo. But we can tell this by what that individual goes on to say and do, and this is in the end not significantly different from the way it works in the case of linguistic predication. Someone using ‘woman’ might also get the wrong end of the stick, might, as is said of the Wittgensteinian pupil, go on in the wrong way, by using ‘woman’ in ways that surprise and confound us.

Even if you accept the above lesson, and come away thinking that there isn't a radical difference between linguistic predication and the use of objects to fulfil that same role, you might well feel that there is still difference enough to call for comment. I couldn't agree more, and it is the business of this section to face up to this fact. Interpreters use ordinary predicate expressions like ‘is a woman’ to display the co-ordination that exists in the use of that word, and I think an interpreter is perfectly justified in using ‘is the ↓sun↓’ to display a similar co-ordination in respect of Romeo's (R). But such co-ordination does not come about miraculously, and there are differences between these two cases.

No one would expect me to explain in a few pages, let alone a paragraph, how it comes to be true that users of linguistic predicates expect, and are right to expect, that their own use of these predicates matches the use made of those predicates by other members of their community. This is because what is necessary to filling in this picture is nothing short of an account of how human beings come to learn or acquire a natural language. Still, I can say something which is anodyne enough to be uncontroversial, but which nonetheless highlights an important difference between linguistic predication and non-linguistic qualification.

Given a certain genetic endowment—one which is probably language-specific, as Chomskyans maintain—we acquire the lexicon of this or that natural language by training in, and/or exposure to, uses of the items which make up that lexicon. More particularly, we all use ‘woman’ in much the same way because we are much the (p.145) same (genetically and in other ways) and we have had something like the same kind of training in or exposure to the use of this word. One of the most obvious differences between linguistic predication and qualification shows up here.

It has been my contention that objects, where this includes individual things, actions, events, states of affairs, etc., are used by human beings to convey information about other objects. In another place, I should like to consider at length whether the genetic endowment which allows us to use objects in this way is part of the same dedicated (or modular) endowment which makes natural language possible. It seems reasonable to think this true; after all, if there is some specialized genetic basis to our abilities to use words for reference and predication, it would be surprising if this didn't also underpin our ability to use non-word objects in these same ways. (Admittedly the referential use of objects is familiar, while their predicational use is not. But I hope that discussion and examples in Chapter 2 show that this disparity is superficial.) Still, whatever may be true of our genetic endowment, it seems obvious that, while we acquire predicate expressions in the lexicon through some kind of training, nothing quite like this happens in respect of objects. Even if I am right about the use of the sun in (R), and in other utterances, I wouldn't claim that we acquire this ability in precisely the same way we acquire the ability to employ the sound ‘woman’. How important is this difference? Does it undermine the whole idea that objects can fulfil predicative functions? While it is easy to anticipate my replies to these questions, I will return to them, after considering a further potential difference between predicates and objects.

I suspect that many would regard expressions like ‘woman’ as relatively immune from the unruly demands of pragmatics. There is a kind of fiction that philosophers of language have often adopted according to which pragmatics makes its contribution, when it does, within certain circumscribed areas of the linguistic realm. Appeal to pragmatics is fine, indeed necessary, for fixing the reference of certain elements in utterances, for tense, and for coping with the potential unclarity that comes from the short cuts we take in speaking, for example, saying ‘the table’ when there are many such things in the immediate and further off environment. But, aside from the often forced examples of radical pragmaticists, the contribution of ‘woman’ to ‘is a woman’, and from thence to the truth conditions of sentences in which the latter predicate operates, does not seem to call on pragmatics.34

In Chapter 1, I thought it best to steer clear of debates about such matters as the semantics/pragmatics distinction (divide?), and I still think this best. But the point that I need to make here doesn't require any heavy-duty theorizing. It is simply this: in so far as you think a predicate expression like ‘woman’ is substantially, or even wholly, a matter of semantics and not pragmatics, you are bound to be disappointed by qualifying uses of the sun. The latter, as with most instances of qualification, are seriously context-dependent. On my account, we need to see the sun as informative (p.146) in order to understand Romeo's utterance, but we must make liberal use of pragmatic considerations in doing so.35

In sum, typically linguistic predicates tend to be acquired by some kind of training, and our dealing with them is rather more semantic than pragmatic, whereas no one thinks we have somehow been trained to use the sun in our understanding of (R), and the context-dependency of the latter is evident and unavoidable. These are real differences—and I will return to the issue of pragmatics and context below—but what I shall suggest now is that they are not as important as they might seem.36 The underlying mechanisms of co-ordination in action and thought—co-ordination that that I called attunement—are really quite similar, whether that attunement is marked by words or objects.

3.9.1. Training and culture

On the simplest model, words are things we are trained to use, and because of this we find it natural to think of the co-ordination marked by words as largely the product of this training. This seems to contrast sharply with the case of objects; it can seem more a matter of luck than anything else when the use of an object as a predicate by one person is taken up in that same way by another. There are two strategies that one might use to soften this contrast.

On the one hand, one could question the role of a narrow and simplistic notion of training in language acquisition. On the other hand, one could insist that there is something very much like training, now understood more liberally, discernible in the many ways we think about objects. Since these strategies are obviously compatible, even reinforcing, I suggest that we adopt them both.

Chomsky argues forcibly that ‘training’, as this notion would ordinarily be understood, is neither obviously present, nor could be efficacious, in our acquisition of language. Wittgenstein ridicules the idea of training as a practice in which a teacher inculcates a naïve pupil's responses by giving explicit instruction. There is no doubt that Chomsky and Wittgenstein come at this issue from very different directions but, taking a wider perspective, one can see a certain sameness in their conclusions.37 What is crucial to both is a conception of what is brought to a putative training session. In their different ways, they subvert the simplistic idea of training by arguing that human beings bring to training what is in effect the very thing that training is supposed to inculcate. In particular, it is argued that the subject of (p.147) linguistic training either possesses language already (in some sense), or comes equipped to respond, perhaps simply by being human, in the ways that the teacher demands.

Even without going into detail, it should be clear how this kind of subversion serves my purposes. Suppose that we use ‘woman’ as we do, not primarily as a result of training, but because we somehow are born with that concept, or, less problematically, that there is a place for some such concept already prepared in our cognitive economy. Then the difference between linguistic and non-linguistic predication will seem vanishingly small. Attunement in our thoughts about objects such as the sun is not something explicitly inculcated: it depends instead on our common humanity, on our reacting to the world in similar ways. But the considerations above suggest that something like an appeal to common humanity is no less necessary for understanding ‘woman’.

The admittedly highly general talk about common humanity introduces the second strand of my softening-up strategy. However, here even more than before, I am forced to make what are only schematic remarks. For the simple truth about a very complicated matter is that the kinds of objects my account calls on in metaphor are typically enmeshed in rich, and often acquired, patterns of thought and behaviour. Objects called on by metaphors are not merely particulars, actions, events, and states of affairs with which we causally interact; they are often bearers of what I earlier rather weakly called ‘cultural’ significance. In ways that are more the concern of anthropologists than philosophers, different societies have somehow managed to turn many of these objects into what are typically called symbols.38 In some cases, for example, that of the sun, the network of symbolic associations is fairly obvious. But in more interesting cases, it is more deeply buried. To take just one example, certain polarities such as that of night and day, or male and female, have had far-reaching, deep, and unobvious effects on the ways in which we classify things that are not themselves straightforwardly related to these (or other) polarities.39

This is no place for even the most potted account of cultural or symbolic significance, nor am I qualified to give one. In any case, spelling out examples of our susceptibility to these significances is not the issue. What is important is understanding how we come to be susceptible in the first place. For my suggestion is that, when one takes on board scepticism about simplistic ideas of training, the way is clear to seeing our acquisition of cultural or symbolic significances as not all that different from our acquisition of linguistic forms. To be sure, I am not denying that there are (p.148) differences; I had better not be, if I am to remain on good terms with Chomsky and his thesis of language-specific genetic endowment. But there is no harm in pointing out that, as in the case of language, we almost certainly end up finding significances in objects because of a potent mixture of hard-wiring (i.e. genetic endowment), sharing of reactions (without explicit instruction) and training, in a more defensible sense, that can be fitted under the rubric ‘culture’.

3.9.2. Significance and context

While still allowing there to be differences between linguistic predication and the use of objects as qualifiers, I have tried various strategies to minimize them, to treat any such difference more as a matter of degree than anything more radical. But for all that I have said, I don't think this is the end of the matter, and I need to confront several obvious further worries, not to say objections, which are in fact interconnected. Doing so will lead directly into some further remarks about context, the real business of this subsection.

As put to me on several occasions, one objection is as follows: language is compositional, productive, and, in these ways, systematic. Part of the point of developing a theory of meaning for a language—part of the point of speaking about a semantics for a language—is precisely to reflect these features. Can I really be saying that the meanings—call them ‘significances’—of objects, when used as qualifiers, share anything like these features?

Consider these sentences:

Juliet is a woman.

George Sand is a woman.

In each case, the predicate expression ‘woman’ makes the same contribution to the truth conditions of the whole sentence, and—as many have argued—the fact that they do make the same contribution is absolutely essential for explaining the compositionality and, hence, the productivity characteristic of natural language. In contrast, the objection continues, consider:

Juliet is the sun,

The warrior Achilles is the sun,

J. M. W. Turner is the sun.

According to the semantic descent account, the sun contributes subtly different things to the truth conditions of each of these sentences, differences that depend in part on the varying contexts in which each is asserted, and in part on the place or places of the sun in our culture. I have spoken of qualification as semantic, but is it really reasonable to think of the vast background network supporting each qualifying use of the sun as admitting the kind of systematicity necessary for compositionality?

The reply to this is not difficult. Of course, there can be no question of the significances of objects being organized in a compositional way. The context-sensitivity of (p.149) objects employed in qualification (where this includes their cultural roles) rules out any such thing. Moreover, if one insists on using ‘semantic’ so that the only things which exhibit this feature compose in the way that words in natural language are reputed to do, then I had better refrain from using this term of our ability to use objects as qualifiers of other objects. However, while these answers suggest something like surrender, this is not so. The lack of compositionality of objects in their qualifying roles is only problematic if it threatens whatever compositionality we think we find in natural language. But, on my account, no such threat exists. Indeed, it is precisely one of the virtues of that account that it leaves the words of natural language—meanings and all—unreformed. Black and theorists like him tell a story about how words, in the context of a metaphor, come to acquire new meanings. It is this kind of project which is threatened by the worry about compositionality, not mine. If ‘woman’, ‘infant’, or ‘sun’ figure in a compositional theory for English, this is well and good. The semantic descent account takes these words with their compositional meanings just as they are, using them only as conduits to the things these words, in their unreformed employment, extend.

As to the term ‘semantic’: while I could just allow that this term is appropriate only of natural language words (and constructions), and then only when they form a system recognizable as compositional, this seems unreasonable and contrary to common philosophical usage. Surely, if someone tells a story using a salt cellar to refer to their car, we would regard that as an exercise of a semantic ability, though we would certainly not wonder whether salt cellars compose. Similarly, if one sees some event as giving us information about another, then, however context-sensitive and unsystematic, we should I think regard what happens, just as we did in the referential case, as an exercise of a semantic ability.

Even if, in respect of acquisition and general functioning, there is no radical difference between words and objects functioning like words, and even if the issue of compositionality is not relevant, there is still a difference that needs further discussion.

One might be willing to accept that there are parallels between words and their meanings, and certain kinds of objects and their significances, but still insist that objects like the sun need pragmatic (contextual) input to avoid a kind of indeterminacy that simply doesn't attach to predicate expressions like ‘woman’. After all, it is one thing to fend off as mistaken interpretations of (R) which, for example, take Romeo to have said that Juliet gets up early, or that she is hot stuff. But the object itself still seems to leave what many would regard as an unacceptably wide range of less off-beam interpretations that can only be reined in by appeal to context. In contrast, though we can entertain the possibility of someone using ‘woman’ in a way which makes us doubt whether it is a woman being spoken about, this has nothing to do with an intrinsic indeterminacy in the use of this predicate expression—an indeterminacy that can only be eliminated by some kind of contextual constraint. Indeed, part of the point of this worry is that ‘woman’ simply doesn't have a point of attachment for context.

There are two ways to respond to this. The first accepts that there is this sort of indeterminacy in cases of qualification, and that qualification depends on (p.150) context in a way that ordinary predication doesn't. But it goes on to insist that, when we understand the nature of the contribution context makes, we should be less inclined to find the admitted difference between predication and qualification to be radical. (I have already made scattered remarks about context, but I will pull them together a bit more immediately below in spelling out this response.) The second response, less concessively, doubts whether this kind of indeterminacy is quite so intrinsic to the use of objects as qualifiers. The claim here is supported by my earlier insistence that there is something misleadingly simplistic about examples like (R) and (T), examples that have carried the burden of exposition in this chapter. (I will expand on this in the next subsection.)

Appeals to pragmatic factors to remove indeterminacy in interpretation can seem ad hoc. This is of course not so when the relevant word, phrase, or sentence is explicitly understood to need contextual supplement. Thus, pronouns, tense, demonstrative expressions, to take three well-known examples, have, as it were, a built-in requirement for contextual specification. Someone simply doesn't understand expressions like ‘he’, ‘was’, or ‘that sofa’ unless he recognizes them as context-dependent; where these items are concerned, no one would think appeal to context is ad hoc. Other words—predicate expressions like ‘is a woman’—do not have what we can think of as ‘hooks’ for the attachment of context. When therefore someone gestures to context to deal with a problematic sentence containing one of these words or phrases, we would seem to have good reason to be suspicious.

I do realize that the distinction between expressions that have hooks for contexts and those that don't is apt to be controversial. However, it is not meant to be a way of reinstating the idea that the only kind of context worth considering is one that can be ‘semanticized’, one which demands explicit parameters onto which contextual information can attach. I can easily imagine there to be expressions demanding a contribution from context, even without this contribution being encoded in the explicit way required for demonstratives, tense, pronominal reference, etc. In any case, the distinction is only intended to be rough and ready. Still, even after these concessions are made, there does seem to be a real difference between the treatment of ordinary predicates, and those in metaphor. There can scarcely be a more obvious candidate for being a predicate expression that is not context-dependent than ‘is the sun’; it is difficult to imagine there being hooks in this phrase that context could attach itself to. Even so, virtually all of the Content Sufficient accounts that I am aware of insist on a role for context in helping us understand the contribution this expression makes to (R). However, unless some story can be told about why we should regard context as relevant to understanding this apparently context-independent predicate, this insistence is little more than a gesture.

As mentioned earlier, there is one writer who recognizes this worry. Stern suggests that we must treat ‘is the sun’ in a metaphor like (R) as subject to an operator (‘M-that’) which turns this context-independent predicate into a context-dependent one. One could say that his operator furnishes the hook for context. A more detailed examination of Stern's view comes in Chapter 5, but it is worth re-emphasizing something here that I said in an earlier footnote. There is something suspect about (p.151) the idea of an operator that imposes context-dependency on a predicate expression that ordinarily gets by without it. If ever the charge of ad hocity was justified, this would be a case. Stern will get his day in court in respect of this charge, but what matters here is not Stern's view but my own. What I shall suggest is that the semantic descent account gives us a way to beat this particular rap: it shows not only why context is important to metaphor, but provides a principled way to see why it is important, a way which doesn't depend just on the fact that metaphors are unintelligible without it.

It should be noted that the above worry about Content Sufficient accounts is not the well-known one about whether it is right to complicate the meaning of predicate expressions in dealing with metaphor. Those who, like Davidson, cleave to an austere account of meaning, have objected to this kind of complexity and have suggested that we don't have a clear idea of how it can be accomplished. But if you reject austerity, and think that the key to metaphor lies in a more sophisticated notion of meaning, you won't be put off that easily. However, the problem with context raised above is not that it makes meaning theory complicated. It is that, while we know that a sophisticated account of meaning needs to take context into consideration, when it comes to predicates like ‘is the sun’ there is no obvious place for context to get a grip.

On the semantic descent account, this whole issue is neatly bypassed. My account is not in the business of generating new meanings for predicate expressions: ‘old’ (austere) meanings do just fine. Thus, in (R), ‘is the (a)’ marks monadic predication, and ‘the sun’ does just what one would expect: it takes us to the object, the sun. However, because it is the object, not the original predicate expression, that is charged with the information-conveying task in (R), there is nothing to prevent our recognizing, and much to encourage, the importance of context in the fulfilment of that task. Indeed, we can say that, in contrast to predicates like ‘is the sun’, objects in their information-conveying modes are thoroughly context-dependent, something true independently of any need to confer intelligibility on this or that metaphor. The mere fact that an object is called upon to serve as a qualifier is by itself reason to count that object as having a ‘hook’ onto which context fastens.

As always, when qualification is in question, it helps to think of cases in which objects take on the other role in the basic combination, namely reference. When someone tells the story of a recent accident using an item on the dinner table to stand for his car, he might begin: ‘The salt cellar is my car …’ Now, the words used in this description are like ‘the sun’ in this sense: these words do not make any explicit demands on context for their intelligibility. To be sure, we need to look to the surrounding circumstances to tell which item is in fact referred to by ‘the salt cellar’, but we need to do this to be au fait with how things are, and not because we would otherwise fail to understand the description. However, while context is not necessary for understanding the descriptive phrase, it is most certainly necessary for understanding the key referential activity that is played out on the dinner table, namely that the salt cellar refers to a certain car. It would be odd to speak of the salt cellar itself as having hooks for the attachment of context, but it is not odd to say (p.152) that the act of using the object in this referential way comes with a built-in demand for contextual help.40

Details aside, the crucial moral of this referential case is that the mere use of an object as a referring device is obviously and naturally understood as dependent on context. There is nothing ad hoc about this dependency; it is not that we find ourselves insisting on the need for context as a result of finding unintelligible the communicative act in which the reference takes place. It is simply that any such use is a use-in-a-context.

This same moral applies to cases of qualification. Independently of metaphor, and of the demand that this or that metaphor be made intelligible, one recognizes the use of objects as qualifiers as a context-dependent activity. Hence, while ‘the sun’ is not obviously bound to any context for its interpretation, when used as a qualifier, the sun most certainly is.

With these remarks, one can better appreciate the real difference between my account's appeal to context and the appeal made by other Content Sufficient accounts. In the usual such account, and Black's can serve as an example here, contexts of the most diverse kinds are ransacked for ‘associated commonplaces’ that then help to make the metaphor intelligible. These commonplaces, if that is ever what they are, are those claimed to be associated with the predicate expression in the metaphor. But as we have seen, the predicate expression in a metaphor is typically not one that has any need of context for its intelligibility; context is thus sort of forced on it. In contrast, in my account, context is tied, not to an expression, but to the object that then comes to serve as a qualifier. Because qualification by objects is a function that is intrinsically context-dependent, there is no forcing of contextual ingredients onto an unwilling recipient.

Here is another way to think about these issues. Ordinary predicate expressions like ‘the sun’ do their predicative work (obviously, in this case, when combined with the copula) largely on their own: context is not necessary for them to function in their communicative or expressive settings. Focusing on this can make objects as qualifiers seem hopelessly indeterminate, because we simply do not think of an object as having a meaning matching that of a predicate expression in natural language. However, this is not the right way to look at qualification by objects. As we have seen, such qualification, like its referential counterpart, is intrinsically (and naturally) context-dependent. So, the right thing to compare with a linguistic predicate is not the object itself, but the object in its context. (Or, to be more precise, we should compare predicate expressions with the object in its contexts. The plural is required because, in some, but by no means all, cases, there can be systematically varying qualificational uses of one and the same object. This is no different from the (p.153) kind of multiplicity of meaning that attaches to natural language predicates such as ‘bug’.)

We should then not think of metaphor as requiring appeal to the object got by descent, the whole of which then requires appeal to context to confer intelligibility on the metaphor. This can make my proposal sound ad hoc in just the way that Black's is. Rather, we should think that intelligibility requires, from the beginning, an appeal to object + context. And, though this will vary with the specific example, many of the parameters of context which serve the use of an object as qualifier are in large part determined by the object itself; this is what it means to take seriously the idea of ‘object + context’. So far then from being ad hoc, we should think of objects as bringing to metaphors certain qualificational potentials. This reinforces a parallel between linguistic predicates and qualification that I have insisted on throughout.

While the above remarks made liberal use of ‘context’, there are distinctions to be made. Once it is recognized that appeals to some kind of context belong with, are required by, each qualifying use of objects, we should separate this kind of context from the more general kind that could figure in any use of a sentence, whether metaphorical or not. (I say something about this below.)

There are two main sources of contextual constraint that focus directly on the use of an object as qualifier, that is, which figure in the formula ‘object + context’. The first is largely linguistic, and the second largely a matter of shared, broadly empirical, knowledge, but this distinction can be blurred in specific cases. On the linguistic side, one begins with the fact that the words from which semantic descent is made exercise some control over the way the relevant object is used. We understand the sun to function monadically because ‘sun’ is embedded in the ‘is a (the)’ schema. (As you will see from the examples discussed in the next chapter, polyadic predication is no more problematic.) Additionally, the qualifying use of the sun is controlled by the fact that that very word is used for this object, instead of one of the many possible co-referential alternatives. The control in this particular example is subtle, and is best brought out by imagining ‘sun’ in (R) replaced by ‘astronomical body at centre of our solar system’, ‘nuclear fireball 193 million miles away’, or even ‘astronomical object worshipped by the Egyptians’. Each of these would transform and disfigure Romeo's (R), even though, in each case, the object got by semantic descent would remain the same. The plain fact, already discussed, is that the sun has a cultural significance for us, though we would be deflected from appealing to this significance if the object were referred to in one of these different ways.

The non-linguistic, but focused, context is made up of the knowledge (and of course beliefs) we have about the object which figures in the metaphor, though these are much too coarse as ways of describing the kind of underlying attitudes we have towards objects that I earlier described as ‘cultural significance’. After all, objects count for us in ways that we often don't notice, and would have difficulty in recovering without help from anthropological friends. Still, in so far as our differential appreciation of objects affects our behaviour and thought, and given the tendency to use ‘knowledge’ even when it is in some sense tacit, there is no harm in speaking of our knowledge of an object's cultural significance.

(p.154) What I prefer to think of as ‘general’ context consists basically of the general factual and linguistic setting within which the metaphorical utterance is made. For example, Romeo doesn't just come out with (R), he says a lot more about the sun before and after the famous line. We also know, in his case, and do generally, lots of things about the speaker and the things spoken about. These include the fact that the subject of the sentence is Juliet, a young woman with whom Romeo is infatuated, that his emotions are in turmoil, etc. Many writers appeal to this kind of general context in connection with metaphor, but the simple truth is that such context is a part of utterance interpretation generally and, though in any given case it might be crucial, it has nothing special to do with metaphor. In fact, though I haven't gone in for the kind of detail necessary to force home this claim, I count it an advantage of the semantic descent account that it allows us to mark distinctions among different kinds of appeal to context. On the one hand, there are contextual ingredients which go with the object of metaphor and help to fix the contribution it makes. On the other, there are contextual ingredients which form the background to the whole of the utterance, and which, for example, help us to disambiguate ambiguous expressions (or objects). There is no doubt that the interplay between these two can make it seem as if there is just one kind of contribution that context makes. But this would be a mistake.

3.9.3. Dispensing with culture

I have used metaphor examples sparingly in this chapter. Romeo's claim about Juliet and the critic's swipe at Tolstoy have carried virtually all of the expository burden, and these two are syntactically almost as simple as metaphors get. While this minimalist strategy has the advantage of not distracting attention from the main lines of my account, it has drawbacks. Although I think we should recognize the role of culturally shared significance in respect of the sun or of infants, these examples are each rather special. It would be wrong to think that my account stands or falls on them, or indeed on the parallel I have suggested between the cultural significances of these particular objects and the notion of word sense. The sun is certainly an object which figures in all sorts of ways in our collective psyche, but that is just the trouble: it is such a flexible symbolic friend that its use in (R) needs heavy-duty support from all three kinds of context, it being one of those cases which give evidence of something like ambiguity. The situation is different with infants, but no more helpful: there can be little doubt that there is a shared stereotype of infantile behaviour, but we think so many other things about infants that context is crucial for triggering this stereotype. Additionally, the very fact that one can speak here of ‘stereotype’ suggests that (T) is conventional, and this might make one wonder whether, as already noted, the metaphor in (T) is dead. The subject of dead metaphor gets a thorough airing in Chapter 4, and I shall also have something to say in that chapter about the more interesting variant of (T) that the critic actually produced. But the point I want to make now is that there are endless examples of metaphor which can lend support to the semantic descent account, and in which context, especially the element of cultural significance, plays a smaller part. This is the less concessive strand of my defensive strategy.

(p.155) Though many more interesting examples will figure in the next chapter, I shall close this one with a taster, something to serve as an encouragement to watch this space. Consider the following claim that might have been made by an observer in the Montague household:

(27) Romeo is an elastic band stretched to its limit.

Or, if you think this too anachronistic, try:

(28) Romeo is an unsecured fifty-pounder on the starboard side with the ship set to come about onto the port tack.41

Unlike (R) and (T), these examples do not depend on being embedded in a rich linguistic context, nor does anything seriously anthropological get a look in. My account requires us to find the object of descent from the phrase ‘elastic band stretched to its limit’, and this is perfectly straightforward. It then requires us to imagine using this object as a qualifier of Romeo. Again, this is straightforward. Even linguistic control of this qualification—control usually exercised by the words from which descent is made—is minimal. We can tell from these words that the relevant object is intended to function like a one-place predicate, but, excuse the pun, there isn't much slack in interpreting what the stretched elastic band tells us about Romeo. From the point of view of interpretability, there is no great difference between someone's saying (27) and someone's saying:

(29) Romeo is so emotionally overwrought that he might suddenly do something irrational.

This is not because the predicate ‘is an elastic band stretched to its limit’ comes to have a new meaning courtesy of a sophisticated theory of meaning; nor is it because there are properties in common between an elastic band in that state and Romeo's state of mind. It is simply because the state of affairs of an elastic band stretched in the way—the very object got by semantic descent from the predicate expression taken in its ‘narrow’ truth-conditional sense—conveys information about Romeo no less efficiently than the linguistic predicate in (29).

The crucial point is that there can be uses of objects as qualifiers which do not require much input from cultural and linguistic context, and which therefore can match purely linguistic predications in respect of determinacy. However, in making this point, I asked you to consider (27) and (29) alongside one another, and I do realize that this is dangerous. It might be taken to imply that these sentences are in some sense equivalent, or even worse, that (29) is a paraphrase of (27). Neither of these implications is intended. With respect to paraphrase: I wouldn't resist treating (29) as a comment someone might make about an assertion of (27); it is of the right form to be what I have called a rationalizing comment—a comment used (p.156) in justifying (27). But this is no more paraphrase than the following elucidatory comment:

(30) An elastic band stretched to its limit can break at any moment, often with painful consequences.

The question of whether (29) is equivalent to (27), and could therefore be a substitute for it, is separate from, and more difficult than, the question of paraphrase. It is easy to imagine someone thinking that the effect of (27), minus only its flourish, could be achieved by (29), and this raises all sorts of questions about the importance, richness, and independence of metaphor in comparison to the literal. However, I certainly didn't intend the introduction of (29) to raise these questions, and I will postpone them to the last section of the next chapter. This should not be taken as a sign of reluctance in dealing with them; it is simply that sensible answers to these questions require more detail about metaphor, and about the semantic descent account, than is on the plate just now.

Notes:

(1) Roger White (1996) is especially good at uncovering examples of this, and he is not reticent about denouncing them. In aiming to avoid this himself, he may perhaps err on the side of an over-rich diet of examples, but his castigation of overly simple accounts is certainly a welcome relief from rather one-dimensional examples and selective discussions. There will be a proper discussion of his account in Ch. 5.

(2) I presume that Quine 1970 is the original source for this term.

(3) I put it this way to remind us that, strictly speaking ‘is the sun’ is the predicate, and ‘the sun’ is only an ingredient in it. As noted in Ch. 2, there is a certain carelessness in usage here. Sometimes philosophers speak of the predicate ‘ewe’ and sometimes of the concept expression ‘ewe’ which figures in the predicate ‘is a ewe’. I think that the latter is correct, but the shorthand is convenient and certainly intelligible. In the present context, it seemed necessary to try to be accurate about this, though as will be shown at length my proposal is not limited to subject-predicate metaphors.

(4) I am perfectly happy to follow Davidson here in speaking of a kind of sharing of properties as characteristic of categorization. But this is not going back on the discussion of these matters in Ch. 2, because I have been careful to resist saying that categorization is explained by or grounded in our noticing that properties are shared or that certain things just seem similar.

(5) More certainly needs to be said about the very idea that we have this kind of reference in (T). Note that the Tolstoy example is one of those mentioned earlier: along with many simple subject-predicate metaphors, it has features of which raise questions about my account which would just not arise in realistically rich and complex cases. Nonetheless, it is useful in other respects, e.g. in this discussion of demonstrative sentences, precisely because of its simplicity.

(6) Only touched on implicitly earlier, what about completing the demonstrative this way: ‘Tolstoy is this kind of thing’? Superficially this sounds alright, but ‘blue’ and ‘kind of thing’ are radically different. Subject to vagueness, which is just not relevant here, ‘blue’ names a property—it effects a division of things into the blue and the non-blue. ‘Kind of thing’ is simply not like this. On the most plausible reading, saying that Tolstoy is this kind of thing might well just be a general way of saying that he is an infant or a human being. On a somewhat forced reading, it may be taken in the way explicitly given in (6) below. If it is, then I take back my claim that we can only achieve what (6) does by twisting the surface syntax of the original.

(7) I have been careful not to mention the views of Josef Stern (1985, 1991, 2000) in my discussion of the demonstrative in (1), though I can imagine that anyone familiar with those views would be puzzled by this omission. Basically, the reason for it is that I want to be fair. His account of metaphor, though it does have more trouble with demonstrative constructions than he seems to realize, seems to me to fall down, not on this relatively technical issue, but on matters more connected with the issue of properties and similarity. I will get to his view in Ch. 5.

(8) It might be thought that (8) should be heard as: ‘Einstein is the brilliant scientist’, where stress plays a semantic role. However, I think that the stressed version is yet another variation, and its availability does not take away from the fact that we have predication and not identity in (8).

(9) Perhaps it is not necessary to say this, but just in case: many identity claims are informative. This is something we knew even before Frege made us realize its importance. But what we have in (8) is not an informative identity, but rather a sentence that seems to be an identity, but is in fact predicational through and through.

(10) I suppose the stressed, ‘Harry is the accountant’ might just achieve predicational status, but it also introduces further complications not relevant to the present point. Note too that it is not necessarily even a consequence of (139) that Harry is an accountant. Though I don't want to get involved here in tangential issues about attributive and referential descriptions, the utterance of ‘Harry is the accountant, but he is not an accountant’, sounds fine to me, and it doesn't take much imagination to see why.

(11) For those who do not find hair-splitting problematic, I add a speculative remark about the reason many do not even notice the ‘ungrammaticality’ of: ‘Ernest is a most awful bore’ or ‘Juliet is a Sun’. There is some ground for thinking that the natural language marker of predication in the simplest subject-predicate sentences is the expression ‘is a’ and not simply the ‘is’. If this is right, then it is not surprising that we find these sentences perfectly alright: it is as if one were saying: ‘Ernest is-a (the) most awful bore’ and ‘Juliet is-a (the) Sun’. Construed in this way, ‘(the) most awful bore’ and ‘(the) Sun’—these very things—become predicates in virtue of their being marked as such by ‘is-a’, and it is therefore not surprising that we do not find real tension between the indefinite article and their definiteness. It scarcely needs saying just how congenial this is to my account.

(12) In this paragraph I have spoken of meaning and meanings, but this is just convenient shorthand, and is most certainly not meant to exclude the Davidsonian truth-conditional account. It is especially important to note this because—whatever some of his supporters have written—Davidson is firm in adopting the non-psychologist stance described in the text.

(13) Semantic descent is clearly a species of reference, albeit one quite different from other members of this large genus. I will say some more about this and about how we can conceive of a determinate, even if non-actual, object in Ch. 4. Also, in that chapter I will finally put to use the conception of an object I have insisted upon throughout—a conception on which events, facts, states of affairs, etc. also count as objects. I mention this here because there should be even less temptation to be worried about the issue of actuality when ‘object’ is interpreted this widely.

(14) My comments about fiction have been brief, but they are important. As the last sentence of this section suggests, even the determinacy of objects got by descent from non-empty predicates calls on our imaginative abilities, our tendency to engage in fictions. Indeed, one way to think of metaphors—something that is encouraged by the semantic descent account—is as small-scale attempts at fiction or, perhaps better, narrative. As I have already noted, this means that one should not get too bogged down in the metaphysics of semantic descent. Creators of metaphor can be understood as telling us very short stories about objects, and their use in characterizing the world. These stories ask us, sometimes but not always, to imagine determinate objects, but they offer no guarantees about their existence, and my account is one about how these objects are used, not about whether we can bump into them.

(15) Nor is it completely uncontroversial that such sameness of saying is preserved in semantic ascent using the truth predicate.

(16) In the more usual case of ascent/descent via the truth predicate, both directions are possible. Yet it can seem from what I have so far said about the metaphorical case that only descent has a point. This is not always so, as will be discussed in the section on dead metaphor in Ch. 4.

(17) Though given the widespread misuse of quotation marks, it may be simply optimistic to regard them as part of everyday practice.

(18) So far from being a problem, the fact that the sun can play different roles in this way seems just that little bit more evidence of the potential of objects to be predicates. If words can be ambiguous, here is evidence that objects can be as well.

(19) Speech act views such as Searle's also see Romeo as making an assertion, but remember that I have classified indirect speech act views as Content Insufficient because content in the relevant sense is not by itself sufficient to explain what Romeo is doing. However, understood correctly, the content provided by the semantic descent account is sufficient for that purpose.

(20) It is also, at least faintly, metaphorical, but I shall let that pass here.

(21) I do realize that, in using Sellars's phrase so soon after citing McDowell, and going so far as to mention Wittgenstein, I am connecting up my discussion to a literature that is becoming ever more vast, thus rendering my few remarks about this nexus of philosophical theses superficial at best. That said, I make no apologies for making the connection, since I think it important to the overall project in this book.

(22) Which is not to say that there is something behind the use of ‘woman’ in (17) and (18) which guarantees that the speaker, interpreter, and ourselves are always going to be in step. The use of the word in these formulas indicates, rather than determines, co-ordination or attunement, so things could go badly wrong, even though they have gone swimmingly up to the point of our acquiescing in this particular interpretative use of ‘woman’.

(23) At the risk of introducing just that bit too much of McDowell, let me try to make the point about attributions, using his notion of ‘sideways-on’ vs direct (i.e. not sideways-on) perspectives. Everything up to now has been mainly about our view of the interpreter's view of Romeo's utterance, all of it therefore sideways-on. Think now about how Romeo might conceive of his utterance (apologies to Shakespeare). He sees the sun on a wonderful spring morning, and finds it to convey information about Juliet; the sun, he reckons (in my terminology) qualifies Juliet. In doing so, he is guided by what he knows about the sun, what he believes, what he believes others believe or know, etc. All of this leads him to make the assertion (R). In doing so, he expects to be understood. Why? Because (as my account has it) his words are straightforward and he expects that his audience will perfectly well understand the kind of background information relevant to this particular qualification. Now imagine the retainer and Fodor approach him: the first says, ‘So you mean she gets up early’, and the second ‘So you mean she's hot stuff’. Here I expect Romeo to be exasperated with the retainer and Fodor, not with himself. From his point of view, he has a perfectly good thought, it was the reason he said what he did, and he would expect that others who share the background information about the sun, and understand him would also understand his utterance. Yet these two characters just didn't get it. Could he have somehow spoken sloppily? Did his words have an ambiguity he should have avoided? How could he put his uncomprehending auditors right? These are all questions from Romeo's point of view—a view that is definitely not sideways-on—but since it is my example, I will take the liberty of imagining some answers from my sideways-on perspective. Romeo did not speak sloppily or ambiguously, no more so than anyone does who uses words that require some sensitivity to what is going on. It was simply that the auditors did not show the requisite sensitivity. Surely, Romeo had a right to think that his auditors knew as much about the sun as he did, and also knew that he was talking about Juliet, the object of his love and devotion. There is a range of things his auditors could have said which might have surprised (even delighted him) because one can often find that one's meaning is best elucidated and expanded upon by someone else. But Romeo is right to think that the retainer and Fodor simply didn't get it. How to help them? Well, he could say some things about the sun by way of elucidation, he could say some things about Juliet by way of providing reasons for his remark. But as for the remark itself, nothing needs to be added: it is not the predicate to which the sun contributed that needs modification, just the auditors. (More on the contrast between the elucidation of a metaphor and a reason for it in the next section.)

(24) Just to remind you: I don't think that metaphors are even mostly subject-predicate in form. For the present, however, I am not questioning this all-too-common assumption, because it makes exposition easier. In the next chapter, it will be shown how the semantic descent account copes with the syntactical variety of realistically complex examples.

(25) Given the central thesis of this book, I am the last person to insist that non-linguistic acts are not expressive and informative. So, take the example in the spirit intended: making a cup of tea might well be expressive, but not always.

(26) If, try as hard as you can, you cannot come up with anything else asserted by Romeo, you are in particular danger of taking sentences like (25) as genuine attempts to put Romeo's thought into words. This is what makes the semantic descent account particularly important in the present context: it gives us a way to preserve the irreducibility of metaphor, while still making space for elucidatory and explanatory comments about particular metaphors.

(27) It might be helpful to contrast the two notions in a context somewhat removed from metaphor. One could think of rationalizations as like the comments that a film director makes (typically on a separate track on the DVD recording) explaining what he hoped to achieve with a particular scene. Elucidations are more like the comments that a film critic might make about how that particular scene achieved whatever effects it did. Note though that there is absolutely no temptation to count either of these as a ‘paraphrases’ of the scene and, though the context is different, the explanation of this is pretty much the same as in the case of metaphor, namely, the fact that the scene itself did not say anything in words.

(28) Though I haven't mentioned this so far, and don't want to make too much of it even now, it really does seem odd to speak of ‘loosening’ our standards of paraphrase (in connection with metaphor) when the notion itself is defined in a way which simply doesn't allow such talk. As noted, any dictionary will tell you, and as is obvious from the word itself, paraphrasing is saying the same thing in other words. While it is clearly reasonable to think that the ‘sameness’ referred to can be tighter or looser, dropping the implication of the phrase ‘in other words’ is not so much loosening the notion as abandoning it. And, given my remarks about commentaries, this is not something we need to do.

(29) Rather surprisingly, Fodor suggested (in an email) that my point about transparency was ‘straight empirical’ and therefore inappropriate as a philosophical objection to other accounts of metaphor. The surprise (to me) is simply that he more than anyone else seems willing to count empirical and therefore contingent theses within philosophical argument. But, more to the present point, he was mistaken in thinking that transparency is any more ‘empirical’ than the fact that metaphors are often taken as assertions, or that we find it odd to be asked to paraphrase a metaphor. For an account of some of the psycholinguistic research on metaphor see Gibbs 1994.

(30) While I have no objection to speaking of ‘ambiguity’, perhaps this is not the best term to use, since what we end up with is closer to polysemy, i.e. multiple related meanings. Or, because semantic descent countenances two kinds of meaning—that of the ordinary predicate-expression and that of the object as proto-predicative—perhaps one should speak of bisemy. Whatever the label, what is important is that semantic descent shows precisely how the meanings are related, and it thereby makes manifest and precise the active role of ordinary meaning in metaphor. This is in contrast to most appeals to polysemy in metaphor, where one finds only hand-waving in the direction of those relations.

(31) An additional weakness of the IA account is what it leaves unsaid. While I allowed it to be intelligible why relevant elucidations and commentaries fit with the deliverances of the IA, this is actually a rather grey area. Without further guidance, it is unclear what we are to do with the images which, according to the IA, confront us. In contrast, though someone might feel that objects do not pin down interpretations in quite the way that linguistic predicates do, it is at least clear what is required of them. (Not that I am suggesting here or in the text that hearers are aware of the semantic function of objects as such. As with the examples given in Ch. 2, we often take objects to be qualifiers without necessarily realizing or labelling what we are doing.)

(32) It is not clear how one could do the experimentation necessary to decide between these alternatives, but a look at certain forms of cross-modal priming (e.g. ‘sun’ priming the sun) might help us decide on the plausibility of the bisemous activation hypothesis.

(33) As you will see, metaphors of whatever syntactical form still display something of a ‘this-is-that’ nature, so it is all too easy to imagine the IA working just as well in these more complex cases. However, this is a mistake. What is at issue in transparency is comprehension of a sentence with a specific form. That there is a useful way we could think about the content of that sentence if we ignore its actual form might well be true, but would be disastrous for transparency.

(34) I have in mind examples discussed in Travis 1981.

(35) One author (Stern 2000) goes so far as to posit a special operator to account for metaphor—an operator which converts ordinary predicates into context-dependent ones. While I shall discuss the problems of this view at some length in Ch. 5, and while I agree with him that context is crucial to metaphor, I can say now that I disagree about the point on which context exerts leverage.

(36) They may well account for what I claimed to be the mistaken idea that linguistic predicates, in contrast to my objects in a qualifying role, have meanings the mastery of which brings about attunement. However, these differences are trouble enough on their own, and I am trying to keep them separate from the myth of meanings that Wittgenstein and others have laboured to expose, and which I discussed in section 3.6.

(37) See Guttenplan 1992 for an attempt at reconciling Chomsky and Wittgenstein, even in the face of Chomsky's resistance to any such thing.

(38) I hope it is obvious that my account of metaphor can be understood as supplying a refinement to the often overly flexible term ‘symbol’. Very roughly, symbols are objects that take on semantic functions. Sometimes, a symbolic object is a simply referential device, but, if I am right, the really interesting case is that in which an object takes on the work of predication.

(39) Here let me acknowledge an enormous debt to Mary Douglas. In her written work, she has made a substantial contribution to our understanding of what is sometimes called ‘analogic’ thinking, a topic which connects directly to my account of metaphor. But, further, I have been fortunate in having been able to have many discussions with her more directly about metaphor.

(40) One might think then of the use of the salt cellar as a bit like a use of a demonstrative: both have built-in demands for contextual clarification. However, I put this remark in a footnote, partly because demonstratives raise many intricate issues that are not relevant here, and partly because I want to avoid claiming that object-reference involves the same kind of explicitness of appeal to context as demonstratives.

(41) Roughly, Romeo is a loose cannon.