In Languages of Art, {1} Nelson Goodman proposes a broad theory of denotation which would embrace all referential functions, including the pictorial, within a single symbolic operation by which one object ``stands for'' another. Within this work, Goodman offers a theory of metaphor as transference---the application of predicates from one domain to objects of another. But one cannot understand Languages of Art without the realization that this work is informed by Goodman's emphatic commitment to nominalism. In this paper I shall first examine Goodman's theory of metaphor within the context of his general theory of symbols in section 1. Next I shall attempt to exhibit a certain tension in Goodman's nominalism between convention and stipulation in section 2. In section 3 , I shall analyze the relationship between nominalism and transference, and conclude that the conventionalist reading of Goodman's nominalism is incompatible with his theory of transference.

Finally, in section 4, I shall suggest that Goodman's account of metaphor is deeply unsatisfactory no matter how it is read insofaras it offers no path to understanding the interpretation of metaphor. For this reason I shall conclude that the exorbitant price demanded by the acceptance of Goodman's theory of metaphor makes this a bargain we should not strike.

Section 1:
Metaphor as Transference.
For Goodman, metaphor must be understood in terms of its role within the general theory of symbols which is Languages of Art. Metaphor arises for Goodman within the symbolic function of expression. Metaphor is present when we say that a picture expresses sadness. Goodman proposes that we understand expression as metaphorical exemplification. Therefore, he must explain both the symbolic function called exemplification and the notion of metaphor which makes exemplification into expression.

Section 1.1:
Exemplification.
As we have said, Goodman begins his examination of expression by analyzing the function of exemplification. Goodman offers an example: ``Before me is a picture of trees and cliffs by the sea, painted in dull grays, and expressing great sadness.'' (50) He distinguishes in this description (1) what objects the picture represents, (2) what properties the picture possesses, and (3) what feelings are expressed by the picture. Goodman suggests that the line between possession and expression is disputable; the situation could have been described by saying that the picture is a sad picture, or that it possesses sadness. In this sense, expression is to be understood in terms of possession, which Goodman thinks of as exemplification. {2} Exemplification, Goodman tells us, is reversed denotation. We say that a picture exemplifies grayness if the label ``gray'' applies to it: the picture is thought of as a sample, like a tailor's swatch, standing in for the range of objects to which the label applies. The picture does not denote the color gray, but is denoted by the predicate ``gray''. When we say the picture exemplifies grayness then, we assert that it is denoted by the predicate ``gray.'' Thus, exemplification must be understood as reversed denotation.

Section 1.2:
Transference.
For Goodman, the judgements that the picture is gray and that it is sad are both reports of what it exemplifies. But the picture is not sad in the same way that it is gray, for a picture literally exemplifies grayness (belongs to the class of gray things) and only metaphorically exemplifies sadness (belongs to the class of things that feel sad). Goodman thinks that it is transference that distinguishes metaphorical from literal exemplification. Metaphorical exemplification is an application of a familiar predicate to a new object. Goodman states that ``applying an old label in a new way . . . is a matter of teaching an old word new tricks.'' (69) Because the label is itself familiar, it carries a history which conflicts with the new application. Thus, for example, we may claim that the picture is sad even though only sentient beings are sad. In this sense, metaphorical exemplification places contradictory demands on our understanding: application of label to object must at once engender sufficient tension to signal novelty and sufficient consonance to make sense. This conflict is central to the operation of metaphor: ``metaphor is an affair between a predicate with a past and an object that yields while protesting. . . . Application of a term is metaphorical only if to some extent contra-indicated.'' (69) From this observation Goodman is able to distinguish metaphorical application both from simple falsity and literal application. If metaphorical application involves both resistance and attraction, simply false predication finds resistance with no attraction, while literal application can be understood as attraction with no resistance. To say that the picture is sad is to re-assign the label; to say it is yellow is to misassign the label; to say it is gray is merely to assign the label. But re-assignment can be misassignment: when we say the picture is gay, the metaphorical application is false.

Goodman explains the difference between misassignment and reassignment in terms of what he calls schemata. He states that labels function as members of families. Goodman thinks of a schema as a group of labels which serves to mark out a group of objects, which he calls a realm. For example, the labels ``yellow,'' ``red,'' ``grey,'' are members of a schema which defines the realm of colored things. Goodman explains that the reassignment which constitutes metaphor involves a change of realm. When a single label in a given schema is applied to an object not in the realm sorted by that schema, the result is a metaphorical predication. Significantly for Goodman, the label's migration to an alien realm is always accompanied by the transposition of other labels from the native schema. Therefore, the use of labels of the old schema in the new realm is organized by the traditional use of those labels in their realm of origin. Thus, for example, the application of the temperature predicate ``warm'' to an element of the realm of hues also determines which hues will be organized under other temperature predicates such as ``cool.''

Described in this way, metaphor can apply to non-verbal, as well as verbal, symbols. The non-verbal metaphor articulated by the judgement that the painting expresses sadness is explained by the series (1) exemplification as the inverse of denotation, (2) possession as exemplification, and (3) metaphorical expression as transferred possession. In this sense, expression is understood as metaphorical possession on the level of non-verbal representation: the sad painting is a case of metaphorical possession of a representational sample, which exemplifies a label coextensive with ``sad''. Thus, Goodman concludes that ``what is expressed is metaphorically exemplified.'' (85)

Section 2:
Nominalism: Practice or Precept?
As I have mentioned, it is impossible to read Languages of Art without noticing Goodman's strong nominalist leanings. It is this nominalism which lies at the heart of Goodman's explanation of metaphor as transference. But there is a curiously persistent equivocation throughout Languages of Art which frustrates the attempt to understand Goodman's nominalism. In this section we shall examine the two competing versions of nominalism which can be discerned in the text in the hope of bolstering our understanding of transference. From Goodman's nominalist perspective, the reason predicates apply is nothing more than ``practice or precept.'' (88) Perhaps the most important effect of this nominalism for Goodman is the ease with which he is able to dismiss the problem of justifying particular metaphorical predications. In opposition to the claims of similarity theorists of metaphor, Goodman emphasizes that the propriety of a metaphorical application of a predicate does not depend on the literal application of the predicate. Does the sad picture have anything in common with the sad woman? Only that the predicate ``sad'' applies to both of them; metaphorically in the first case and literally in the second. For Goodman, ``the question why predicates apply as they do metaphorically is much the same as the question why they apply as they do literally.'' (78) And this is a question Goodman's nominalism will not permit him to answer.

For Goodman, labels are grouped together into schemata, and the union of the extensions of the labels in a schema constitutes a realm. The connection between a label and its extension takes the form of a set of rules of association. But here Goodman's nominalism becomes ambiguous, for what he takes to be the source of such rules is unclear.

Throughout Languages of Art, Goodman equivocates between a stipulative nominalism which would make the rules of association nothing more than precepts handed down from unknown sources and a conventional nominalism which would root these rules in conventional practice. Thus, practice at once orders and is ordered by conditions for the applicability of a label:

The choice among systems is free; but given a system, the question whether a newly encountered object is a desk or a unicorn-picture or is represented by a certain painting is a question of the propriety, under that system, of projecting the predicate ``desk'' or the predicate ``unicorn-picture'' or the painting over the thing in question, and the decision both is guided by and guides usage for that system. (40--41) In this sense, the rules of association play two opposing roles. The result of this tension is that various of the mechanisms of transference are thrown into question. In his account of metaphor as transference, Goodman separates the question of the truth of a predication from that of its literalness. An application of a predicate to an object is literal just in case the object is a member of the realm associated with the schema of which the label is a member. An application is metaphorical if the object is not of the realm conventionally picked out by the schema of the label, and therefore a new set of rules of association between schema and a foreign realm must be created. An application is truthful if the label applies to the object under the rules of association in force, and otherwise is false. But if the rules of association linking schema to realm are ordered in part by convention, then the question of truth of a predication begins to creep closer to that of its literalness.

The significance of this tension becomes clear only when we recognize the importance of nominalism for Goodman's theory of metaphor. Because Goodman will not allow the possibility of any eidetic essence in labels, the rules of association comprise the only mechanism by which Goodman can explain the tension and resolution of the re-assignment involved in transference. Similarly, Goodman has recourse only to these rules in distinguishing the misassignment of false attribution from the re-assignment of metaphorical attribution. But the tension between practice and precept in Goodman's nominalism makes the power of his theory to account for these details dubitable. Therefore, we shall have to consider more closely the plausibility of the importance given to convention in Goodman's nominalism in section 3.

Section 3:
Convention and Metaphor.
In the last section we saw how Goodman's nominalism is driven by the dual forces of convention and stipulation. In this section I shall examine critically the ability of convention to accomplish the numerous tasks set before it by Goodman's theory. For each of these tasks, I shall argue that convention alone is insufficient. What is needed to complete the explanations is an explicit appeal to the semantic content of predicates---an appeal forbidden by Goodman's nominalism. It would seem that conventionalist side of Goodman's nominalism is incompatible with his account of metaphor as transference. For this reason we shall conclude that Goodman can only save his theory of metaphor by insisting upon a purely stipulative nominalism whose rules of association are not shaped by convention.

Section 3.1:
Objections.
An immediate consequence of nominalism on Goodman's theory of metaphor is that the transference requisite for expression can take place between any two realms. Because there is no essence connecting a predicate to its extension, there is no possible theoretical reason to oppose either the separation of the label from its native realm or the reapplication of the label to objects of a foreign realm. Indeed, metaphor is all the more compelling when the transference is particularly striking: Metaphor is most potent when the transferred schema effects a new and notable organization rather than a mere relabeling of an old one. Where the organization by an immigrant schema coincides with an organization already otherwise effected in the new realm, the sole interest of the metaphor lies in how this organization is thus related to the application of the schema in its home realm, and sometimes to what the labels of the schema exemplify. But where an unaccustomed organization results, new associations and discriminations are also made within the realm of transfer; and the metaphor is the more telling as these are more intriguing and significant. (79--80) But in making metaphor dependent on convention in this way, Goodman's account remains unable to capture the appropriateness of certain applications of a predicate over others. This theory has an easy explanation for the possibility of metaphor; it is simpler to violate convention than essential affinities between objects and labels. But this ease in breaking the bonds between a label and an object of its native realm and in establishing new ties between that label and objects of a foreign realm corresponds to a difficulty in distinguishing one combination from any other. In particular, it is difficult to see how Goodman's account could explain why some applications seem more fitting and others less. Surely

[(a)] The picture is sad

has something to recommend it over

[(b)] The picture is bovine.

Does the former fly in the face of convention less than the latter? Then, according to Goodman, (a) should be a less compelling locution than (b). But if (b) is more compelling because of its obscurity, why does it remain so rare and (a) so common? Would speakers adopt (b) to convey the thoughts they currently express with (a) if they only discovered it? No, the difference between (a) and (b) consists in something more than their conformity to convention; and this something more is unreachable from Goodman's nominalist position. It seems to me that the difference in conventionality between the two is testament to, rather than the cause of, the relative ``fittingness'' of (a).

A related challenge to the conventional nominalist program concerns the impotence of convention to answer to the contrary demands of metaphorical exemplification. As we have indicated, if applicability depends on convention, the difference between truth and literalness is obscured. For transference to occur, the application must be novel but fitting, strange but satisfying. But to say that an application is novel is to mark it as unconventional; how then can it be fitting for the conventional nominalist? Conversely, if an application is appropriate, then it is conventional; for a conventional nominalist to speak of it as surprising is nonsense. It seems then, that the tension which drives transference is impossible to explain within the conventionalist aspect of Goodman's nominalism.

Even if Goodman could supply explanations for these problems, his conventionalism would face another obstacle in the notion of a failed metaphorical predication. As we have seen, Goodman believes in the possibility of false metaphorical application, encountered in the judgement that the picture expresses gayness. But assuming the process of re-assignment makes sense, it is hard to see how any assignment could fail outright. If, as we have seen, transference is possible between any two realms, what reason could there be to oppose a particular sorting? On the conventionalist reading of nominalism, what is there to distinguish misassignment from re-assignment? Both flout convention, or else they would be simple assignment (literal truth). If misassignment is more unconventional than re-assignment, why does it not become a more dramatic re-assignment, as Goodman suggests? Once again, Goodman's nominalism prevents him from defending the distinction in terms of the semantic features of predicates, which seems the most natural way to mark the difference.

Section 3.2:
Goodman's Defense.
Goodman mounts a brief defense against these charges by retreating to the stipulative side of nominalism. He notes that the picture exemplifies any and all properties it expresses, whether or not anyone brings them to our attention by uttering the relevant predication: ``Sad'' may apply to a picture even though no one ever happens to use the term in describing the picture; and calling a picture sad by no means makes it so. This is not to say that whether a picture is sad is independent of the use of ``sad'' but that given, by practice or precept, the use of ``sad'', applicability to the picture is not arbitrary. (88) If, as this passage suggests, all predications can be judged against stable rules of association regarding the applicability of the relevant predicate, then we might resuscitate Goodman's distinction of re-assignment from misassignment. For, assuming there is a matter of fact about the applicability of a predicate to an object whether or not anyone has thus applied it, then a misassignment could be understood as a simple violation of this matter of fact. In contrast, re-assignment could be understood as a novel application of a predicate which does in fact apply. Similarly, this solution might be used to explain the tension required of transference. Because transference involves reassignment, tension would be introduced in the conflict between the novelty and the genuine propriety of the predication.

But this defense presupposes that there is a matter of fact about the applicability of predicates to an object even before they have been applied to it. Whether or not this presupposition is ultimately true (I think it is), it constitutes a flagrant violation of conventional nominalism. Thus, the defense Goodman proposes to save his theory can serve its purpose only if he is willing to surrender the conventionalist convictions which endangered the theory in the first place. Needless to say, Goodman refuses to make this concession, and therefore his defense is unsuccessful. What, we ask, does Goodman mean when he asserts that the picture is sad although no one has seen fit to utter (a)? To remain faithful to conventionalism, Goodman can only answer that a picture is sad if it is a member of the class of things denoted by the label ``sad'', which is true of the things this label conventionally applies to. To say, then, that the picture is sad is to say that the picture is a thing which conventional practice calls sad.

Once again, Goodman rejects the possibility that there is an essence common to sad things, on the basis of which convention might proceed. If we were allowed such a notion of essence, we would say that convention endorses the application of ``sad'' to an object just in case that object has the essence common to all sad things. However, without such an essence, it is unclear to me that convention has any judgement to offer about the applicability of ``sad'' to the picture if no one has ever attempted the combination. What is convention if not the history of applications up to the present moment? If this history includes no instances of the application of a predicate to an object, and if we may not appeal to past applications of the predicate to other objects, on what basis can history sanction the new predication?

On the other hand, if Goodman would give up conventionalism and embrace a fully stipulative nominalism, this defense would be completely successful. For according to the stipulational reading, the regulations which bind label to extension are simply created by fiat. On this picture, a label applies to an object just in case the rules of association allow the combination. This strategy could rescue both the distinction between misassignment and re-assignment and the tension central to transference. But Goodman refuses to abandon convention:

Since practice and precept vary, possession and exemplification are not absolute either; and what is actually said about a picture is not always altogether irrelevant to what the picture expresses. . . . Establishment of the referential relationship is a matter of singling out certain properties for attention, of selecting associations with certain other objects. Verbal discourse is not least among the many factors that aid in founding and nurturing such association. (88) Despite the guarded tone of this passage, Goodman clearly wants it both ways---rules of association determine and are determined by practice. Simply, this defense is unsuccessful because it does not go far enough in rejecting the conventional nominalism so inimical to Goodman's picture of metaphor as transference.

Section 4:
The Absence of Interpretation.
Goodman offers a powerful theory of metaphor in Languages of Art . His account neatly binds together verbal and non-verbal metaphor and incorporates the notion of metaphor within a larger theory of symbols. Moreover, Goodman uses this functional analysis of metaphor to classify the tropes of hyperbole, understatement, overemphasis, underemphasis, and irony (Languages of Art, section II. 8). But in spite of these theoretical advantages, I have little sympathy for Goodman's account because it remains silent with respect to the most important aspects of metaphor. However serious the threats posed to Goodman's theory of metaphor by our considerations in section 3, these objections against the technical feasibility of his program are far less important than the recognition that even if the project were successful, it would fail completely to address the hermeneutics of metaphor. Ultimately, what is most conspicuously absent from Goodman's account of metaphor is any conception of what it means to say that the picture is sad. While his distinction of metaphorical from literal exemplification as a signaled deviation from the operative rules of association may explain our recognition of (a) as metaphor, this explanation does not speak to the interpretation of (a). But (a) does not merely flout rules of association: it also articulates content. Not only does Goodman fail to provide any theory of the content of (a), he emphatically renounces the most promising path toward such a theory in stressing his commitment to nominalism.

It seems to me that any account of the meaning of a metaphorical application of a predicate must depend on the meaning of a literal application of that predicate. {3} Indeed it is precisely this dependence which distinguishes metaphor from simple polysemy: the independence of the two readings of

[(c)] The money is in the bank.

prevents us from calling (c) metaphor. But Goodman repudiates this distinction in insisting that the sad woman and the sad picture share only the condition of being classified under the label ``sad''. With this renunciation, he retreats absolutely from the possibility of explaining the content of predication. The result of this nominalism is a permanent inability to say what it is that (a) means.

In divorcing predication from anything other than adherence to and deviation from rules of association, Goodman demands that we abandon all hope of interpreting (a). But if Goodman is content to leave the question of why predicates apply ``to the cosmologist,'' (28) I remain unsatisfied with this response. For if predication only registered varying levels of conformity to rules of association, it is hard to imagine what might constitute motivation for the articulation of a predication.

The separation of predication from content in Goodman's theory removes all incentive from the projects of discourse and analysis. In this respect, I believe that his account is utterly inadequate. The appeal of metaphor lies in its ability to suggest questions and provide insight. The observation that metaphorical predication is deviant is but the initiation of a much longer process. Metaphor is not only innovation but also representation, not merely poiesis but also mimesis. In this spirit Ricoeur asks:

Why should we draw new meanings from our language if we have nothing new to say, no new world to project? The creations of language would be devoid of sense unless they served the general project of letting new worlds emerge by means of poetry. . . {4} In short, Goodman's account of metaphor demands that we abandon all aspirations to a hermeneutics of metaphor and content ourselves with mere recognition of the phenomenon. What Goodman offers as rewards for this sacrifice are a rapprochement between verbal and non-verbal metaphor and a tidy tropological taxonomy. To my mind, these make paltry recompense for our loss.

                                    Jonathan Cohen / jdc5@cs.uchicago.edu