Catherine 2.
Elgin
3. Metaphor and Reference"'
Metaphors abound and are often understood without diffi-
culty. Theories of metaphr also abound. Yet an understanding
of how metaphors function is surprisingly hard. I take meta-
phor to be a semantic matter: Terms genuinely denote the ob-
jects they metaphorically describe. And literally false sentences
may be genuinely true when construed as metaphors. Meta-
phorical sentences have the same cognitive functions as literal
sentences and others besides. O n e such additional function is
to liken their literal and metaphorical objects.
Denotation
Application of old words to newly encountered objects is a
staple of language use. Often we proceed without difficulty.
We unhesitatingly apply the term "car" to unusual passenger
vehicles, and the term " o d d " to unfamiliar integers not evenly
divisible by two. Where habit or stipulation settle the refer-
ence, where we have no inclination to withhold a term, its ap-
plication is literal.
Metaphor also involves applying old words to new objects.
But here in metaphor matters are not so straightforward. For
the habits and stipulations that guide literal usage both suggest
and preclude metaphorical applications 1 . People do not belong
* This paper derives substantially from Chapters 4 and 8 of my With Refer-
ence to Reference, Indianopolis: Hackett Publishing Company, 1983.
1 Cf. N. Goodman, Languages of Art, Indianopolis: Hackett, 1976, pp. 81-85
54 Catherine Z. Elgin
to the literal extensions of botanical predicates. Nevertheless,
the demise of the follower of Richard II is correctly described
as follows:
The weeds which his broad spreading leaves did shelter,
That seemed in eating him to hold him up,
Are plucked up root and all by Bolingbroke.
(.Richard I I , III, 4)
Richard literally had no sheltering leaves, nor had his flatterers
roots. Yet much as weeds weaken the plants that shelter them
and disfigure the garden they infest, Bushy and Green, while
under Richard's protection and apparently supporting him,
undermined his reign and damaged his realm. To avoid the
further damage they would do if left unchecked, Bolingbroke,
like a gardener, had them destroyed. Because we know how to
characterize plants as weeds, we know how to use " w e e d " to
characterize people. The metaphorical interpretation of a term
thus depends on its prior literal interpretation.
The priority is semantic, not just historical or etymological.
It is not only because the characterization of plants as weeds
antedates that characterization of people that the latter is
deemed metaphorical. Semantic priority concerns current us-
age. The application of " w e e d " to uncultivated plants that
weaken the soil, deprive other plants of nutrients, and eventu-
ally take over a garden is what justifes its application to
Richard's flatterers. If neither interpretation recalled and de-
pended on the other, then regardless of their etymology, their
current usage would be non-metaphorical.
Ordinarily, the sorting of objects effected by a term's meta-
phorical application reflects the sorting effected by its literal
application. This does not account for metaphorical uses of
Active terms, though, for they sort nothing. Whatever I am
getting at when I say that Oswald is an ogre, I am not (even
metaphorically) consigning him to the null set.
and "Metaphor as Moonlighting", Of Mind and Other Matters, Cam-
bridge, MA: Harvard University Press, 1984, pp. 71-77.
3. M e t a p h o r and Reference 55
Before I can say how fictive terms function metaphorically,
I need to say a little about how they function literally. Since
such terms have, and are known to have null denotation, their
denotations do not distinguish among them. Instead, as Good-
man argues, we should explicate them on the basis of certain
terms that denote them. These terms are replacements of the
schema: p-description2. Thus "Hamlet", "the Prince of Den-
mark", "the melancholy Dane", and the like are all Hamlet-
descriptions; "Peter Pan", "the boy who never grew up", "the
leader of the lost boys", and the like, all Peter-Pan-descrip-
tions; and so on. Understanding a fictive term involves recog-
nizing it as an instance of particular p-descriptions; and creat-
ing a "fictive individual" like Hamlet, or a "fictive kind" like
ogres involves bringing a variety of expressions to instantiate
the same p-descriptions.
Just how we identify diverse items as instances of a particu-
lar predicate is a difficult question. But there is no reason to
think that it is any more difficult to explain how we recognize
that certain linguistic items instantiate particular p-descrip-
tions than it is to explain how we recognize that certain vehi-
cles instantiate "car" or "truck". Nor is there any reason to
think that the explanations of the two cases differ significantly.
Upon reading Peter Pan, we immediately and directly recog-
nize certain expressions as Peter-Pan-descriptions and others
as Captain-Hook-descriptions. We do not need to map our
terms onto their referents to ascertain that our classifications
of them under various p-descriptions are correct.
2 Cf. Languages of Art, pp. 21-26. Goodman introduces "p-description" to
disambiguate "description of p". Under one interpretation, a description of
Peter Pan describes a denizen of Never Never Land; under another, it de-
scribes no one, there being no such person as Peter Pan. O n the former
interpretation "description of p" is a schema of a one place predicate that
classifies expressions on the basis of kind. O n the latter, it is the schema of a
two place predicate that links a term with its denotation. I follow G o o d -
man in restricting "description of p" to its two place semantic interpreta-
tion, and using "p-description" to classify terms directly.
56 Catherine Z. Elgin
P-descriptions sort literal uses of Active terms. Their meta-
phorical uses reflect these sortings 3 . If Oswald is an ogre, some
predicates that count as ogre-descriptions - "fearsome", "bru-
tal", and "nasty", perhaps - apply denotively to Oswald. But
no single literal predicate need underwrite the metaphorical
transfer. Applying " o g r e " metaphorically to Oswald may
bring about a likening that no literal predicate precisely cap-
tures.
A Active characterization such as "Simon Legree is an ogre"
may involve a fictive metaphor. Since neither "Simon Legree"
nor " o g r e " denotes, the metaphor effects no reclassification of
denotata. But a reclassification of descriptions occurs. "Ogre-
description" applies metaphorically to characterizations that
"Simon-Legree-description" applies to literally.
A non-null term may occur metaphorically in a fictive con-
text as well. In " D o n Juan is a wolf", the predicate " w o l f " lit-
erally denotes wolves and metaphorically denotes obsessively
amatory men. Accordingly, it metaphorically picks out obses-
sively-amatory-man-descriptions, including Don-Juan-de-
scriptions. To say that Don Juan is a wolf is then to effect a
metaphorical classification of Don-Juan-descriptions. The
metaphorical use of a term effects a new classification not only
of the objects it denotes, but also of associated descriptions
and pictures.
Metaphor is not just ambiguity. Like an ambiguous term, a
metaphorical term has more than one extension. But while the
several interpretations of an ambiguous expression are seman-
tically independent, a term's metaphorical application depends
on its literal application. Although some tokens of " w e e d " ap-
ply to unwanted flora and others to mourning garb, neither
application influences the other. "Weed" then is ambiguous as
between botanical and sartorial readings. But because some to-
kens of " w e e d " apply literally to unwanted plants others apply
metaphorically to Richard's flatterers. The role of " w e e d " in
3 N . Goodman, Ways of Worldmaking, Indianopolis: Hackett Publishing
Company, 1979, p. 104.
3. Metaphor and Reference 57
classifying people recalls and is guided by its role in classifying
plants. To apply "weed" to people then is to use the term
metaphorically.
A metaphor, Goodman says, is "an affair between a predi-
cate with a past and an object that yields while protesting" 4 .
Metaphor depends on both attraction and resistance. A term
resists its metaphorical application for the objects it applies to
typically do not belong to its literal extension. But it attracts
that application, for it likens the objects in its metaphorical
extension to those in its literal extension.
Systems
Taken in isolation, a predicate is simply a device for grouping
objects. Seldom are predicates taken in isolation. They usually
function collectively as families of alternatives that sort the ob-
jects in a domain. Such a family may be called a scheme, and
the objects it sorts its realm. The odd/even scheme applies lit-
erally to the mathematical realm; a color scheme, to the visible
realm; the animal/vegetable/mineral scheme, to the physical
realm. A system is a scheme applied to a realm5.
N o more than literal predicates do metaphorical ones func-
tion in isolation. Whether explicitly or tacitly, an entire
scheme typically transfers to a new realm, the reassignment of
predicates reflecting their assignments in their native realm.
Metaphor thus is a displacement of a term or scheme under the
influence of the rules and habits that determine its original ap-
plication.
Metaphors that merely supply new labels for old kinds are
coextensive with available literal labels. Thus, for instance, we
might classify alternatives in a binary system as " o n " and " o f f '
or as "pass" and "fail". Such metaphors simply point up or
exemplify features their literal and metaphorical referents
4 Languages of Art, p. 69.
5 Ibid., pp. 71-74.
58 Catherine Z. Elgin
share. But such cases are comparatively rare. Usually a meta-
phor effects a sorting that no literal scheme does. The objects
that instantiate the same metaphorical predicate then coinstan-
tiate no literal predicate.
Schemes organize the objects in their realms by establishing
relations of inclusion and exclusion. These relations continue
to obtain when the scheme transfers. The entire scheme need
not, of course, be explicitly mapped onto the new realm. In
metaphorical as well as literal cases, semantic systems are sys-
tems of implicit alternatives. If we label a social program a war
on poverty, we introduce a network of terms for characteriz-
ing responses to social conditions. With some, we are at peace.
A m o n g these are allies in the current struggle (for example,
public education), as well as potential foes (perhaps industrial
pollution) we are not yet ready to take on. O u r various at-
tempts to alleviate poverty can be described as campaigns, bat-
tles, and skirmishes, and be evaluated as victories or defeats.
This need not be explicit. Simply calling our program a war
makes the descriptive resources of the militaristic scheme
available for describing the social realm.
Metaphor does not require an alien realm. A scheme may
apply to its native realm in such a way as to effect a novel
sorting. In labeling someone an "intellectual midget", we em-
ploy a scheme that literally classifies people by stature to clas-
sify them on the basis of cognitive attainments. A new sorting
of the realm results, since height is no measure of epistemic
achievement.
Whether a token is literal or metaphorical depends on how
it is interpreted. Acceptable interpretations can classify the
same tokens differently. Under an interpretation that con-
strues the extension narrowly, relatively few applications are
literal, relatively many metaphorical. Under one that construes
extension more broadly, more applications are literal. A n in-
terpretation that counts only animal appendages "legs" takes
all talk of table legs metaphorically. According to a more gen-
erous construal for animals and furniture alike fall under the
literal application of "leg".
3. M e t a p h o r and Reference 59
Where tokens have divergent extensions, interpretation de-
cides whether the difference is a matter of metaphor or ambi-
guity. If the application of a term to objects in one domain
precedes and informs its application to objects in the other, the
latter is metaphorical. If the applications are semantically inde-
pendent, the expression is ambiguous. It is not always obvious
which is the case, and different acceptable interpretations may
yield different verdicts. Whether "lame duck" is ambiguous or
metaphorical depends on whether, according to the interpreta-
tion in effect, politicians who are about to leave office exem-
plify features of disabled water fowl.
Truth
Truth is not exclusively literal. Thinkers who relate everything
to a single central thesis, Isaiah Berlin contends, are hedgehogs
and those who pursue a variety of unrelated cognitive ends,
foxes 6 . Even though "Plato is a hedgehog" is literally false, un-
der Berlin's system it is metaphorically true; whereas "Plato is
a f o x " is both literally and metaphorically false. " T r u e " and
"false" then are not equivalent to "literally true" and "literally
false"; nor is "denotes" equivalent to "literally denotes". For
"hedgehog" metaphorically denotes Plato.
To construe metaphors as true involves no violation of
Tarski's theory 7 . Any interpreted extensional language admits
of a Tarski truth definition, so long as its primitives are finite
in number and certain types of self reference are barred. A
truth definition is formally correct if it yields no paradoxical
sentences such as
This very sentence is false.
6 I. Berlin, The Hedgehog and the Fox, N e w York: Simon and Schuster,
1953, p. 1.
7 Cf. A. Tarski, "The Concept of Truth in Formalized Languages", Logic,
Seamrttics, and Metmathematics, Oxford: Clarendon Press, 1956.
60 Catherine Z. Elgin
It is materially adequate if the sentences it yields are replace-
ments of the schema:
(T) X is true if and only if p
where the name of a sentence replaces X and the sentence itself
(or its translation into the metalanguage) replaces p. Interpre-
tation of primitives is given trivially, by a list. Thus "snow"
denotes snow; "white" denotes each white thing; and so on.
Nothing in Tarski's formal definition of truth mandates any
particular choice of primitives or any particular assignment of
objects as their extensions. Any semantic system that satisfies
the formal requirements has a truth definition.
Metaphorical systems typically do so 8 . Metaphor consists in
applying a scheme to a new realm, or in applying it to its old
realm in a new way. Any scheme that has a truth definition
when interpreted literally has a truth definition when inter-
preted metaphorically. Whether snow is identified with frozen
particles of water vapor, or with interference on the television
screen, "snow" denotes snow. And "Snow is white" is true if
and only if snow is white. Since "true" is a semantic predicate,
a truth definition can be given only for an interpreted lan-
guage. But whether the interpretation is literal or metaphorical
makes no difference.
The construction of a truth definition explains what it is for
the sentences of a system to be true, not what it is for them to
be metaphorical. A complete account of metaphorical truth
plainly requires both. The point here is that if we accept
Tarski's criterion for a formally correct and materially ade-
quate definition of truth, we have no reason to deny that many
metaphorical sentences, even though literally false, are true.
Each system has its own truth definition. Under a literal
system, "Plato is a hedgehog" is false; under a metaphorical
one, it is true. The fact that we define "truth" for separate sys-
8 For an argument that mataphor is extensional, cf., C. Z. Elgin and I. Schef-
fler, "Mainsprings of Metaphor", The Journal of Philosohy, 84 (1987): 331-
335.
3. Metaphor and Reference 61
tems rather than for the language as a whole is no special cause
for concern. Tarski recognized that as it stands a natural lan-
guage does not admit of a truth definition. Regimentation is
necessary to eliminate paradox-generating semantic terms, and
to partition the vocabulary into primitive and non-primitive
terms. Multiple, divergent definitions are required to handle
vagueness and indexicality. And so on. Moreover, natural lan-
guage is evidently too unwieldy to accommodate itself readily
to the demands of regimentation. Truth definitions for sublan-
guages are far easier to supply.
Recognition of this often leads to treating a preferred sub-
language (usually, the language of science) as though it were
the whole language, thereby effectively excluding all other sen-
tences as insusceptible of truth. The cost of this move is pro-
hibitive. Not only metaphorical sentences, but also metaphysi-
cal sentences, evaluative sentences, statements of propositional
attitude, and much of ordinary non-scientific discourse then
do not admit of truth. Since the formal requirements for a
truth definition dictate no such exclusion, and since we are
strongly inclined to consider such sentences true, it is better to
define "true" for the systems or sublanguages such sentences
belong to rather than conclude that because they fail to fit into
a preferred regimentation of the language of science, they are
not candidates for truth.
Davidson 9 contends that a truth definition yields semantic
rules for a language. Because he thinks no such rules can be
given for metaphors, he denies that metaphors can be true10.
But the "rules" a truth definition yields are trivial, and are
equally satisfied whether a sentence is interpreted literally or
metaphorically. So if a truth definition gives semantic rules,
there is no reason to deny that there are semantic rules for
9 D. Davidson, "Semantics for Natural Language", in: D. Davidson and G.
Harman (eds.), The Logic of Grammar, Encino: Dickinson, 1975, p. 18.
10 Except, of course, metaphorical sentences like "No man is an island",
which are true under their literal interpretation. Cf. D. Davidson, "What
Metaphors Mean", in: S. Sacks (ed.), On Metaphor, Chicago: University of
Chicago Press, 1979, pp. 29-45.
62 C a t h e r i n e Z . Elgin
metaphors. Such rules, like the rules for literal language, will of
course be utterly uninformative.
To avoid paradox, we need not banish all semantic terms
from the object language. Only those coextensive with "true",
"false", and "denotes" and the like need be excluded. Since
"literal(ly)" and "metaphorical(ly)" give rise to no paradox,
they may remain. If the object language retains them,
"S is p" is metaphorically true = S is metaphorically p
just as
"S is p" is literally true = S is literally/).
If the object language excludes them, we are left with the origi-
nal formulation of Convention (T):
" 5 is p" is true = S is p.
But this is indifferent as between literal and metaphorical read-
ings-
Perhaps Davidson is concerned with the rules for primitive
denotation, not just with sentences that satisfy Convention
(T). But the same point applies to these. If the object language
excludes "literal(ly)" and "metaphorical(ly) " , that
" s n o w " denotes snow
is indifferent as between literal and metaphorical readings. If it
includes them,
" s n o w " literally denotes literal snow
and
" s n o w " metaphorically denotes metaphorical snow.
The so-called rules of primitive denotation remain trivial. We
no more know how to identify the stuff " s n o w " metaphori-
cally applies to on the basis of the latter rule than we know
how to identify the stuff " s n o w " literally applies to on the
basis of the former.
3. Metaphor and Reference 63
Metaphor as Cognitive
Davidson further contends that metaphor is non-cognitive.
This popular canard is hard to fathom given the contributions
metaphors to indisputably cognitive disciplines.
Computer metaphors configure the domain of cognitive sci-
ence. These metaphors, Boyd argues 11 , cannot be dismissed as
merely decorative or heuristic, for we currently have no non-
metaphorical way to draw the same distinctions. The constel-
lation of computer metaphors organizes the psychological
realm in a way that no literal scheme does. Until a coextensive
literal scheme develops, the metaphor is ineliminable.
The metaphors serve to focus psychological research. The
goal of that research is not to discover what the metaphor's
author intended, but how human cognition unfolds - some-
thing no more known to the author than to anyone else work-
ing in the field. In science, a metaphorical claim functions like
any other hypothesis. Diverse members of the scientific com-
munity articulate, clarify, disambiguate, and elaborate it. Such
a metaphor does not reside in a single work, nor is it the prop-
erty of a single author. Like a literal hypothesis, a metaphorical
one will be incorporated into a scientific theory if it proves
fruitful, explanatory, and (at least approximately) true.
Metaphorical scientific claims evidently perform the same
functions as literal ones. Both are open to intersubjective
scrutiny. Both can be contested, confirmed or disconfirmed by
evidence, accepted and incorporated into a science or rejected
as false or as trivial or as lacking in explanatory power. The
computer metaphor, with its talk of inputs, outputs, accessing,
and processing, facilitates communication and verbal reason-
ing about human cognition. The metaphor then both organizes
the phenomena for investigation and provides a vocabulary in
which to conduct that investigation. Surely a metaphor that
plays these roles enhances understanding.
11 R. Boyd, "Metaphor and Theory Change", in: A. Ortony (ed.) Metaphor
and Thought, Cambridge: Cambridge University Press, 1979.
64 Catherine Z. Elgin
Metaphor functions cognitively in another way as well. Re-
call that a scheme determines a system of kinds according to
which the objects in a realm are classified. No scheme provides
a label for every set of objects in its realm. One function of
metaphor is to introduce a new mode of organization by clas-
sifying under a single label objects that are classed under di-
verse literal labels12. Metaphor thus augments our conceptual
repertoire. It enlarges our stock of predicates for classifying
the objects in a domain. This might, of course, be done with-
out recourse to metaphor. We could simply introduce a new
literal taxonomy to effect novel groupings. The problem is to
discover what new groupings are wanted. The metaphorical
application of a scheme to a new realm employs a mode of
organization that has proven useful elsewhere. By deploying
available conceptual resources, the metaphor facilitates com-
prehension and communication about the newly organized
realm. At the same time, it exhibits affinities between its literal
and metaphorical realms.
Likening
The metaphorical application of a term likens the objects in its
metaphorical and literal extensions. The difficulty is to say
how.
Metaphor is often said to be grounded in similarity. A term
then applies metaphorically to certain objects because they are
similar to the ones it applies to literally. The problem is that
everything is somehow similar to everything else. So if similar-
ity underwrites metaphor, every term applies metaphorically
to every object. This just is not so. In Richard II, "weed" ap-
plies metaphorically to Bushy and Green, not to Bolingbroke,
Richard, or John of Gaunt. Unlike similarity, metaphorical
likening is selective.
12 Cf. "Metaphor as Moonlighting".
3. M e t a p h o r and R e f e r e n c e 65
T h e metaphorical application of " w e e d " to B u s h y and
G r e e n does not stem f r o m their satisfying s o m e general for-
mula
S is metaphorically a weed s S is literally p
where p stands for s o m e finite string of specifiable predicates
that literal weeds instantiate. F o r metaphors typically lack lit-
eral paraphrases. N o r does the application of a term to diverse
extensions automatically effect likening. W h e n a term applies
a m b i g u o u s l y to objects in different extensions, no likening
takes place.
It is tempting here to try to pass the buck. W h y not say that
metaphorical likening results f r o m s o m e sort of psychological
association? T h e n " w e e d " applies metaphorically to B u s h y
and G r e e n because calling t h e m weeds p r o v o k e s an association
with undesirable plants. Since calling B o l i n g b r o k e a weed elic-
its no such association, he is not in the metaphorical extension
of the term. A n d since n o n e is p r o m p t e d when m o u r n i n g
clothes are called weeds, that both plants and clothing are
called weeds is mere ambiguity. This, to be sure, does not solve
the p r o b l e m . B u t it does s u p p l y an excuse f o r denying the
p r o b l e m is ours. If metaphorical likening is a matter of p s y c h o -
logical association, it is a matter for psychologists to explain.
O n e does not have to be particularly pessimistic a b o u t
p s y c h o l o g y ' s prospects to find this m o v e unattractive. If the
application of the term " w e e d " to B u s h y and G r e e n induces a
psychological association while its application to B o l i n g b r o k e
o r to m o u r n i n g garb does not, it is reasonable to expect seman-
tic differences to account f o r these psychological differences.
So even if the exact nature of the association is a matter for
psychologists, the differences between metaphorical and non-
metaphorical applications of a term seem not to be. M y thesis
is this: Metaphorical likening is b r o u g h t about b y chains of
reference linking literal and metaphorical extensions of a term.
T h e tenability of m y thesis plainly depends on what I take
chains of reference to be.
66 Catherine Z. Elgin
Like Goodman, I recognize two basic modes of reference -
denotation and exemplification. Denotation goes from a label
to the objects that label applies to. Exemplification goes in the
opposite direction - from a symbol to labels that apply to it. If
" g r e e n " denotes grass, "green" refers to grass; if grass exempli-
fies "green", grass refers to "green". Exemplification is not,
however, the converse of denotation. To denote an object, a
term need only refer to it. To exemplify a label, an object must
both refer to and instantiate that label. Only if grass is green
can it exemplify "green". Exemplification is but a subrelation
of the converse of denotation. It is the mode of reference that
connects with what they sample. So it can transform an object
with no history of symbolizing - a rutabaga or a rhinocer-
ous - into a symbol 13 .
N o t every symbol is purely denotational or purely exempli-
ficational. Some symbols are referentially complex, involving a
number of interacting referential functions. Metaphors are
such symbols. But they are sufficiently complex that before
considering them it will be fruitful first to investigate the se-
mantics of allusion.
Allusion is a form of referential action at a distance. One
thing alludes to another by referring to it indirectly. I might,
for instance, allude to my joblessness by mentioning that I
have lots of free time, a condition exemplified by people who
are out of work. The simplest chains are these: 1) a alludes to b
by denoting something c that exemplifies b; and 2) a alludes to
b by exemplifying something c that denotes b. Schematically,
the two simplest cases look like this:
In more complicated cases, referential chains are longer: a
alludes to b by denoting or exemplifying something that in
turn denotes or exemplifies something that ... denotes or ex-
emplifies b 14.
13 Languages of Art, pp. 50-71; With Reference to Reference, pp. 71-96.
14 Cf. N. Goodman, "Routes of Reference", Of Mind and Other Matters, pp.
55-71.
3. Metaphor and Reference 67
a b c
c a b
Figure 1
The intermediate links of a referential chain typically are not
explicitly identified. If an allusion is at all subtle, the length
and configuration of its referential chains and the identities of
their intermediate links will not be obvious.
Metaphorical likening, I suggest, is a matter of allusion. A
metaphor then denotes the objects in its metaphorical exten-
sion and alludes to those in its literal extension. Let me illus-
trate:
An epidemic of essentialisra spread throughout philosophy.
"Epidemic" metaphorically denotes essentialism which exem-
plifes a label such as "disorder acquired by contact with those
who have it", a label that also denotes literal epidemics. The
chain then goes from the term to its metaphorical denotation,
from there to a label both literal and metaphorical denotata
exemplify, and on to its literal extension. Schematically:
label
term exemplified
metaphorical literal
extension extension
Figure 2
Joint exemplification of the labels in question need not ante-
date the metaphorical application of the term. A metaphor can
effect a likening by bringing certain shared characteristics to
68 Catherine Z. Elgin
the fore — that is, by making the objects in its extensions exem-
plify labels they previously only instantiated.
"Hume's Treatise is a gold mine" is more complex. Typi-
cally, "gold mine" applies metaphorically to sources of great
wealth or profit. Still, the rewards of the Treatise are hardly
financial. Rather, "gold mine" metaphorically denotes Hume's
Treatise, which exemplifies a label such as "very rewarding".
Sources of wealth that "gold mine" ordinarily applies to meta-
phorically also exemplify this label. These in turn exemplify
"highly profitable", a label literal gold mines likewise exem-
plify. Thus:
"gold mine' "very r e w a r d i n g * "highly profitable'
sources
of wealth
Figure 3
In such a case, the chain of reference involves a previous
metaphorical application of the term.
How are we to identify the vehicles of metaphorical liken-
ing? Sometimes there are multiple candidates. Suppose Mavis
calls Travis a troglodyte. Travis and the individuals in the lit-
eral extension of "troglodyte" might jointly exemplify a num-
ber of labels - "stupid", "loutish", "primitive", "subhuman".
Individually and in combination these labels underwrite the
metaphor. So the effectiveness of one chain of reference does
not preclude others. Metaphors often are secured by multiple
chains.
Travis, of course, is not literally subhuman. So a full explica-
tion of how "troglodyte" functions metaphorically requires
explaining how "subhuman" does. The relevant chain of refer-
ence might look like this:
3. M e t a p h o r and Reference 69
'primitive
label'
/\
/
"metaphorically "literally
"troglodyte* subhuman' subhuman'
Travis humanoids
Figure 4
Multiple referential chains evidently may connect meta-
phorical and literal extensions of a term, and their links may
involve further metaphors. Moreover, regardless of the refer-
ential chains we have already identified, there remains the
prospect others. So although a chain of reference can explain a
particular affinity between metaphorical and literal referents,
no chain or group of chains can claim to constitute a full literal
paraphrase. For we are never in a position to maintain that
known chains exhaust the metaphor.
Often, as we've seen, a metaphor effects a novel sorting of
the objects in its domain. The objects it groups together con-
stitute the extension of no literal term. Once we recognize the
way chains of reference link metaphorical and literal referents,
the explanation is straightforward.
In the simplest cases, the metaphorical referent of a term
exemplifies some of the same labels as its literal referent 15 . But
not every collection of objects has a literal predicate associated
with it, so there need be no literal term that applies to just the
objects in the metaphorical realm that exemplify labels in
question. If there is not, the application of the metaphor brings
about a novel sorting.
15 To avoid needlessly complicating my discussion, I consider only the sim-
plest cases. The point is the same for more complex ones.
70 C a t h e r i n e Z. Elgin
An example may be helpful here. Consider
Fred is a mere coat rack.
Fred and literal coat racks jointly exemplify "functions mainly
to keep coats off the floor". Fred, we may conclude, does lit-
tle - his primary contribution to the greater good being to
hold coats. Certainly there are literal predicates that character-
ize such people - "ineffectual", "inconsequential", "of little
account", and so on. But not everyone who instantiates these
literal predicates, indeed, not even everyone who exemplifies
them, exemplifies "functions mainly to keep coats off the
floor". So not everybody in the extension of these literal predi-
cates is in the extension of the metaphor. If "a mere coat rack"
effects a novel sorting of the human population, no simple lit-
eral term denotes all and only those people who are metaphor-
ically mere coat racks.
An especially apt metaphor may evoke a shock of recogni-
tion. Even though its object does not belong to its literal ex-
tension, its application to that object seems peculiarly appo-
site. Such metaphors sometimes evoke aftershocks as well. Be-
sides the primary chains of reference from their literal to their
metaphorical applications, we find chains that go the other
way - from the metaphorical back to the literal. The metaphor
then prompts a reconsideration of term's literal referents as a
result of their being likened to its metaphorical referents. Con-
sider:
If they don't surrender, we'll turn their country into a parking lot.
The primary chains of reference go from parking lot construc-
tion to the enemy nation's destruction. "Turn it into a parking
lot" applies metaphorically to the country because, unless it
surrenders, we will cause it to exemplify such labels as "lev-
eled", "demolished", "flattened" - labels exemplified during
parking lot construction. But that the annihilation of an enemy
nation admits of metaphorical description as turning it into a
parking lot provokes a reconceptualization of urban renewal.
Secondary chains of reference then go from the destruction of
3. Metaphor and Reference 71
countries to the construction of parking lots. These chains in-
volve joint exemplification of a different group of labels - la-
bels like "devastated", "ravaged", "laid waste". The metaphor
then underscores the fact that parking lot construction causes
the terrain to exemplify these labels.
The lore of our fathers, Q u i n e is wont to say, is a fabric of
sentences - perhaps a more intricately woven one than he
imagines. A term that denotes one thing alludes to something
else. It may exemplify as well. It has numerous metaphorical
extensions whose members exemplify some of the same labels
as its literal referents, and others whose connections to its lit-
eral referents are more remote.
The complexity and interconnectedness of the term's vari-
ous referential functions explains why, when a strand breaks,
the fabric does not unravel. Typically, a strand is multiply
tethered. Severed in one place, elsewhere it holds fast. More-
over, to relieve the tension when the fabric becomes strained,
we can cut the strand at any of several places. And in different
cases different types of alteration seem best.
When cognitive psychologists discover that their literal
claims are at odds with their computer metaphor, they may
sacrifice literal assertions to safeguard metaphorical ones. Then
the metaphors can guide their quest for a better literal account.
When we can no longer accept a sentence as literally true,
we can sometimes still hold it true by construing it metaphori-
cally. Religious assertions that contravene scientific findings
often admit of such reconstruals.
When ethologists discover that lions are cowardly and goril-
las gentle, we can safeguard earlier metaphors and allusions by
reconfiguring referential chains - tethering them to lion-de-
scriptions and gorilla-descriptions, false ones as it turns out,
rather than to actual lions and gorillas.
The multiplicity of referential connections results in a cer-
tain latitude for choice. Investigation may force us to conclude
that some modification in a referential network is needed -
that a given term does not refer in all the ways we thought.
Still, empirical investigations do not pinpoint a unique source
72 Catherine Z. Elgin
of difficulty or determine what form modification has take.
Our understanding of the term's several symbolic roles, an
evaluation of their relative importance and centrality, and an
appreciation of the human enterprises that embed them
ground our choice among revisions. If these choices are not
uniquely determined by the evidence, neither are they ad hoc.
