Sei sulla pagina 1di 5

Comparing Two Views of Metaphor Semantics

Carl Vogel? and Josef van Genabithy


Trinity College?
University of Dublin
Computer Science
Dublin 2, Ireland
vogel@tcd.ie

Dublin City Universityy


Computer Applications
Dublin 9, Ireland
josef@compapp.dcu.ie
Abstract
Criteria are presented and argued as necessary qualities of any formal semantic account of the meaning of metaphorical
sentences in natural language. Two recent frameworks that attempt to provide semantic analyses for metaphor are
critiqued relative to these criteria.

1 Introduction
One line of thought about metaphor is that it is a defective form of language use. Certainly, it has been neglected
by most researchers in natural language semantics working within a model theoretic tradition. This is in part because by its very nature as nonliteral language, metaphorical sentences are hard to accommodate in any model of
sentence meaning that gives prominence to truth conditions. Davidson (1984) has gone so far as to suggest that
metaphoricity isnt a property of sentences themselves,
but uses of sentences, something that takes them outside
the field of study for semanticists. However, there has
been a recent approach in semantics that gives equal attention to the change of information state induced by the
interpretation of a sentence (hence its impact on the interpretation of subsequent sentences). Such an approach
makes it possible to address some aspects of language use
in the formal semantic models.
There are two approaches that we know of that attempt to offer a compositional semantics for metaphor.
The first (Vogel, 1998, 1999) assumes translation from
natural language to a formal logical language. Vogel (1998)
uses translation to first order logic and Vogel (1999) uses a
more expressive intensional logic. The logical languages
are given nonclassical interpretations, however; a dynamic
system is used so that a nonliteral expression can have the
impact on subsequent discourse that though literally false,
it is nonliterally true. The second approach is offered by
van Genabith (1999). This system uses a fully classical,

but expressive type theoretic framework. In this work,


each metaphorical sentence is read as corresponding reduced simile.
A goal of this paper is to critique these two approaches.
We outline a number of features that should be present in
a good model theoretic semantics for metaphorical sentences. We provide an overview of each of the two main
approaches in this area, evaluating them with respect to
the desiderata. Finally, we compare the theories directly
with each other.

2 Desiderata
2.1 Compositionality
On the one hand, it is often assumed that metaphor presents
a prima facie case for semantics to deal with a sentence
whose meaning is more than the sum of its parts, on the
other hand, it is argued that a metaphor is simply not effectively paraphrased by a related simile. Davidson (1984)
makes this point quite clearly.1
Both the elliptical simile theory of metaphor
and its more sophisticated variant, which equates
the figurative meaning of the metaphor with the
literal meaning of a simile, share a fatal defect.
They make the hidden meaning of the metaphor
all too obvious and accessible. In each case the
hidden meaning is to be found simply by looking
to the literal meaning of what is usually a painfully
1 Pg.

254.

trivial simile. This like thatTolstoy is like an


infant, the earth like a floor. It is trivial because
everything is like everything, and in endless ways.
Metaphors are often very difficult to interpret and,
so it is said, impossible to paraphrase.

It is important to realize that Davidsons criticism only


applies if a corresponding simile is allowed to range into
the trivial. It is more than doubtful, however, that in most
cases of actual use similes are used trivially in communication. In addition, it remains to address the question of
what that hidden meaning that is more than the sum of
the parts could be. A theory of metaphor that provides an
answer should at least capture the fact that there is not just
any arbitrary likeness between the vehicle and tenor that
a simile seems to explicitly existentially quantify away.
The metaphorical meaning must be given a clear model
just as literal predicates are modeled by their characteristic sets. Moreover, this meaning must emerge from the
composition of the parts of a sentence in a context in a
way that discriminates between metaphorical and literal
readings.

2.2 Truth Conditions


Another way of stating the above constraint on theories of
metaphor is that truth conditions remain important to our
understanding of language. On the face of it, metaphors
and similes have very different truth conditions, and this
difference should be localized in any theory of metaphor.
Similes cannot be false; they assert the existence of a likeness that does not have to be pinned to any particular
property (1).
(1)

The nation is like a cheesecake.


It follows that the negation of a simile is false:

(2)

The nation is not like a cheesecake.

2.3 Aptness
The aptness of a metaphorical sentence is something like
a measure of its appropriateness in a given context. However, it has a different nature than choice of speech act or
other pragmatic decisions that occur in the formulation of
sentences to determine appropriateness.
(6)

Could you please pass the salt?

(7)

This soup could do with a bit more salt.

(8)

This soup was not made with the waters of the Great
Salt Lake.

This is partly evident from the fact that speech acts


combine with metaphors. The difference to finding the
right level of politeness is that aptness is a measure of
semantic appropriateness, even though it does draw on
world knowledge, and context in addition to the sentence
at hand. We use the expression semantic appropriateness to address the intuition that while literally the sentence seems false (or true, depending on the negative polarity), relative to the nonliteral senses at stake, the sentence is true. In fact, the primary issue at stake with aptness is that although the grounds need to be relativized to
nonliteral senses, truth does seem to be the right way of
characterizing the basic level of aptness of a metaphor.
Of course, this is not all there is to aptness. Some
are more apt than others. This seems to involve ancillary predications, and the degree to which the metaphor
extends or falls apart under scrutiny. The contribution
of these additional predications is still in terms of their
holding or not, but with some further assessment of their
number, and as well of their analogical importance to the
metaphor at hand, where analogical importance can be determined structurally using mappings between associated
implicative complexes (Veale and Keane, 1992).2

Notice, however that this observation only holds if


we are prepared to admit trivial likeness as in Davidsons
. . . everything is like everything, and in endless ways.
Trivial use of similes of this sort is probably quite rare
in actual communicative exchanges (except in puns and
jokes). Rather, as is the case with metaphor, the communicative function of a simile is an instruction for the
recipient to establish some particular properties instantiating likeness between tenor and vehicle.
By contrast, metaphors seem to be literally false (3),
and their negations true (4).

2.4 Novelty

(3)

The whale is a blimp.

(4)

The whale is not a blimp.

It is an accepted metaphorical sense of (10) that allows


the sentence to be used to express that Leslie is clever.

(5)

No man is an island

Yet, sentences like (5) demonstrate that metaphoricity


transcends falsity. It isnt the case that every false sentence is metaphorical.

Clearly metaphoricity depends on language use rather than


timeless properties of sentences. There is a difference between the first use of a new metaphor and the re-use of
an established one. First uses are rather more difficult
to process as they involve essentially rendering another
sense for an existing predication.
(9)

Leslie washes books.

(10) Leslie is a fox.

2 By implicative complex we mean just the set of sentences that


pertain to some expression the set of sentences implied by an expression (when its a sentence on its own), the set of sentences that imply it,
etc.

However, in a context in which (9) is literally false, additional information is required to make the sentence interpretable. There are any number of scenarios that render the sentence metaphorically apt, one of which is that
Leslie copy edits manuscripts of books. Once that extension is made, it is readily possible to re-use that sense.
The first use can be quite difficult to obtain, and it is
claimed that there are syntactic constraints on where new
metaphors can be constructed that are distinct from constraints on re-use of existing metaphors (Vogel, 1998).
(11) Sal washes newspaper articles.
Moreover, it is necessary to account for the fact that
eventually, novel metaphors become so entrenched in everyday use that they cannot be discriminated from expressions whose interpretation is intended to be literal. By
now fox as nonliterally used in (10) is close to being
a dead metaphor. Dead metaphors are ubiquitous in language. It is necessary to have a succinct model of the potential for transition from novel use to accepted metaphor
to dead metaphor.

2.5 Semantics-Semantics Interface


Important to a theory which satisfies the above constraints
is that it integrate the interpretation of metaphor and other
nonliteral language (whose context change potential and
truth conditions are of primary interest) with the interpretation of literal language. While the different uses of language can be discriminated in some sense within an overarching system. It is inappropriate to have one mechanism for literal language and an altogether different module for dealing with nonliteral language. The reason for
this is suggested by the preceding section: it has often
been demonstrated that language coordination isnt often
by explicit negotiation but by use and mimicry (Garrod
and Anderson, 1987; Healey, 1995). This (and popular perceptions that prescriptive linguistic institutions ultimately fail) suggests that the primary mechanisms for literal language ontogenesis are closely tied to the creation
and death of metaphor. If literal language springs from the
nonliteral, then it is the literal that is deviant if either is.
Deviance is to be the wrong mode of expression here:
it is more satisfying to have a single framework in which
both literal and nonliteral language are handled equally
but discriminated in the important senses than to set up
systems in which semantics is done on literal language
and only some other processing on nonliteral language.

3 A Critique of Recent Proposals


We contrast two recent proposals for integrated semantic
analysis of literal and nonliteral sentences in formal semantics.

3.1 Intensional Dynamic Semantics


3.1.1 Overview
Vogel (1999) presents an intentional predicate calculus
as a meaning representation language for both literal and
nonliteral expressions. The language has greater than first
order expressivity in that it admits predicate polysemy,
where each sense of a predicate corresponds to a corresponding word sense. This is a very weak extension of
first order logic, though, as it amounts to being able to
index predicates to enumerate their meanings ( BANK1 is
a financial institution, BANK2 is the effluvia of a body
of water, ....). It is a real extension, nonetheless, as it
requires a more complex semantics that provides an index for the characteristic set of individuals satisfying the
predicate in each of its senses. Some indices are classified as literal, and others are classified as nonliteral. Indices can be referred to directly within the language, but
reference to indices isnt essential for syntactic or semantic well-formedness. Underspecification of indices simply results in ambiguity. New indices can be created, and
new tuples can be added to their characteristic sets. Vogel (1999) provides one set of constraints on this process
that restricts where characteristics can be extended. Once
a characteristic set is extended, the newly resulting model
provides the context of interpretation for subsequent discourse. Thus there are two ways in which a classical
predicate logic is extended. The first is to address polysemy. The approach here is to make the system weakly
intensionaldirect reference to indices is possible in the
language but not necessary; no constraints are placed on
the multiplicity of senses. The second issue is that the
system is given dynamic interpretation. The interpretation function must be examined both before and after the
interpretation of sentences. Some sentences have the capacity to extend the function, and this extension is passed
on to the interpretation of subsequent discourse.
3.1.2 Critique
Compositionality. As presented by Vogel (1999), this
system treats metaphor as a mode of expression distinct
from simile. Metaphoricity comes from distinguishing
the sense of the predicate. In a way, this reduces metaphoricity to polysemy with a preference ordering on senses of
expressions: new uses are most metaphorical, nonliteral
but old ones admit new tuples into their remit, old uses are
deemed literal. The system presumes a translation function that maps natural language sentences into the very
rich meaning representation language and at the point of
translation introduces ambiguity: no stipulations are given
on how parts of natural language syntax guide the interpretation into sense selecting expressions or deixis. However, once translated, the sentences is fully distinct from
related simile. The hidden meaning of the metaphor is
just the pointer to the particular sense at stake. In this
framework, metaphorical meaning is thus modeled exten-

sionally.
Truth Conditions. In this system a predicate can be
false relative to one sense and true relative to another,
whether those senses are classified as literal or nonliteral.
Whether the exact truth conditions proposed are correct is
subject for debate.
Aptness. The theory says nothing about aptness, apart
from supplying a way for the metaphorical sentence to be
nonliterally true, and reducing that truth to extensionality. What the system lacks is a way of clarifying what
other predicates require extension, given that one has extended one. It is clear that the sort of mechanism that
would work in this context is one that examines the web
of implications that a predicate participates in, and on the
basis of structural analogy maps those predicates to predicates corresponding to the analogical domain. More has
to be said about this interface, but it is clear that this exactly the sort of theory that has been worked out by Veale
and Keane (1992) in the context of network encodings
rather than first order encodings.
Novelty. Ontogenetic uses of metaphor are distinguished
from re-uses of existing senses for predicates. Shift of
senses is modeled by reclassification of indices as literal
or nonliteral, however no constraints are placed on this
mechanism besides the basic classification. The precise
syntactic (predicate form) and semantic (literal falsity of
the predicate) constraints on first use of a metaphor generate empirical claims that require testing.
Semantics-Semantics Interface. Because all senses of
expressions are extensionalized as characteristic sets of
predicates at indices, literal and nonliteral expressions are
handled within the same overarching framework, with a
difference in precise function corresponding to the classification of indices at stake as literal or nonliteral.

3.2 Type Theoretic Semantics


3.2.1 Overview
In van Genabith (1999) a standard type theoretic approach
is employed to provide an analysis of metaphors as reduced similes (cf. Davidson (1984) and discussion above).
On this account (10) is translated as Leslie and the set
of all foxes share a common property. Sense extension
is accounted for indirectly in that the common property
includes at least the extension of Leslie and the set of
all foxes. However, this simple-minded approach is easily trivialized by the universal property (the property of
being identical to oneself). The translation is strengthened by requiring that the property sought be non-trivial.
The type theoretic representation is obtained compositionally (Montague, 1973). The approach captures both
direct copula constructions and more complex metaphors

involving other predicates (such as sentence (9)). A Prolog implementation is provided.


3.2.2 Critique
Compositionality. Compositionality is one of the motivating factors for this approach. van Genabith (1999)
provides a syntactic fragment and corresponding semantic interpretations for each rule in the fragment. Each sentence generated by the fragment is ambiguous between
a literal and a metaphorical translation which interprets
a metaphor as a corresponding, reduced simile, as discussed above.
Truth Conditions. The truth conditions are classical.
The system derives its power from the translation. Given a
metaphor, the corresponding simile translations augmented
with a non-triviality constraint come out as contingent (either true or false).
Aptness. Like the approach of Vogel (1999), this framework relies on the existing literature on metaphor recognition to determine what counts as a good metaphor. The
approach offers an initial idea towards simile property instantiation (i.e. property resolution). It remains to be seen
whether this scales up to more complex examples.
Novelty. This theory was formulated with compositionality and interface to other semantic modules as its primary concern. It does not address issues of fossilization. The system is like classical logic in being completely static. The re-use of metaphor isnt distinguished
from first use, but is distinguished from literal language.
Semantics-Semantics Interface. The formulation offered
by van Genabith (1999) satisfies this desideratum because
that was it motivating concern. The same mechanisms underly the literal and nonliteral interpretation of sentences.
What discriminates the two interpretations is precisely the
syntax semantics interfacethe choice of translation rules.

4 Discussion
This abstract offers a set of criteria for a semantic theory
for metaphorical uses of sentences. We have described
two theories that offer compositional denotational semantics to metaphors, something that other theories of metaphor
interpretation lack.

Acknowledgements
Many thanks to Julia Hockenmaier for last minute assistance.

References
Donald Davidson. What metaphors mean. In Donald
Davidson, editor, Inquiries into Truth and Interpretation, pages 24564. Oxford: Oxford University Press,
1984.
Simon C. Garrod and Anthony Anderson. Saying what
you mean in dialogue: A study in conceptual and semantic co-ordination. Cognition, 27:181218, 1987.
Pat Healey. Communication as a Special Case of Misunderstanding: Semantic Coordination in Dialogue. PhD
thesis, Centre for Cognitive Science, University of Edinburgh, 1995.
R. Montague. The proper treatment of quantification in
ordinary english. In J. Hintikka e.a., editor, Approaches
to Natural Language, pages 221242. Reidel, 1973.
Josef van Genabith. Metaphor and type theory. Symposium on Metaphor, Artificial Intelligence and Cognition, Society for Artificial Intelligence and Simulation
of Behavior, April 6-9., 1999.
Tony Veale and Mark Keane. Conceptual scaffolding: A
spatially founded meaning representation for metaphor
comprehension. Computational Intelligence, 8(3):494
519, 1992.
Carl Vogel. A dynamic semantics for novel metaphor.
Third International Conference on Information Theoretic Methods in Logic, Language and Computation,
Hsi-Tou, Taiwan, June 16-19., 1998.
Carl Vogel. An intensional dynamic semantics for polysemy, sense extension and metaphor. Symposium on
Metaphor, Artificial Intelligence and Cognition, Society for Artificial Intelligence and Simulation of Behavior, April 6-9, 1999.

Potrebbero piacerti anche