metaphorik.de 06/2004

In this article we argue that abstract inferential principles based on Gricean maxims or, even more radically, on a unique principle of relevance cannot adequately account for how interlocutors actually proceed in inferring utterance meanings. We advocate an intermediate level of inferential principles—metonymies—that are, on the one hand, abstract enough to be used as inferential schemata and, on the other hand, have enough specific conceptual content to serve as guideposts in utterance interpretation. We define conceptual metonymy as a contingent, i.e. non-necessary, relation within one conceptual domain between a source meaning and a target meaning, in which the source meaning provides mental access to the target meaning. We regard such metonymic relations as multipurpose conceptual devices not restricted to language but used in other semiotic systems and thinking as well. Furthermore, we argue that in a prototypical metonymy the target meaning is conceptually more prominent, i.e. more in the focus of attention, than the source meaning. Prototypical metonymies not only make target meanings accessible but also available, e.g. as new topics, for further elaboration in the ensuing discourse. Metonymies in this sense are ubiquitous as conceptual tools in natural language. They function on the referential, predicational and illocutionary levels of speech acts, and they organize conceptual content in the lexicon, interact with grammatical structure, and play a key role in the ad hoc creation and understanding of pragmatic meaning.

In diesem Artikel argumentieren wir, dass abstrakte Schlussprinzipien, die auf Griceschen Maximen oder sogar nur auf einem einzigen Relevanzprinzip beruhen, nicht angemessen erklären können, wie Interaktanten Äußerungsbedeutungen tatsächlich erschließen. Wir plädieren für die Existenz einer unterhalb dieser abstrakten inferenziellen Prinzipien angesiedelten Ebene von metonymischen Schlussprinzipien, die einerseits allgemein genug sind, um als Inferenzschemata zu dienen, aber andererseits auch einen hinreichend spezifischen Gehalt haben, um als “Wegweiser” für die Erschließung von Äußerungsbedeutungen zu fungieren. Wir definieren eine ‘konzeptuelle Metonymie’ als eine kontingente, d.h. nicht-notwendige Beziehung zwischen einer Ursprungsbedeutung und einer Zielbedeutung innerhalb einer konzeptuellen Domäne, wobei die Ursprungsbedeutung den mentalen Zugang zur Zielbedeutung erleichtert. Wir betrachten solche metonymischen Beziehungen als flexible kognitive Werkzeuge, die nicht nur in der Sprache, sondern auch in anderen Zeichensystemen und im Denken Anwendung finden. In einer prototypischen Metonymie ist die Zielbedeutung dominanter, d.h. mehr im Fokus der Aufmerksamkeit, als die Ursprungsbedeutung. Prototypische Metonymien ermöglichen nicht nur den mentalen Zugang zu Zielbedeutungen, sondern stehen, beispielsweise als neues Thema, zur weiteren Bearbeitung im nachfolgenden Diskurs verfügbar. Metonymien in diesem Sinne sind in der natürlichen Sprache als konzeptuelle Prozesse allgegenwärtig. Sie manifestieren sich auf der referenziellen, prädikativen und illokutiven Ebene, und sie strukturieren das Lexikon, interagieren mit der Grammatik und spielen eine Schlüsselrolle in der Produktion und dem Verstehen pragmatischer Bedeutungen.

1. Metonymic reasoning and pragmatic inferencing

Conceptual metonymy is a cognitive process that is pervasive in grammar, the lexicon, conceptual structure, and language use. Metonymies provide what we call natural inference schemas (Thornburg & Panther 1997) that guide much of pragmatic reasoning in the construction of meaning, especially in the determination of explicit meaning, i.e. explicature, and implicit meaning, i.e. generalized and particularized conversational implicature (see e.g. Gibbs 1994, 1999; Levinson 2000).

Relevance theorists (e.g. Sperber & Wilson 2002) and cognitive linguists (e.g. Lakoff 1987, Fauconnier & Turner 2002) emphasize that the cognitive processes operative in the interpretation of communicative acts are usually entirely spontaneous and automatic. The rational reconstruction of these processes shows indeed that pragmatic meanings are conceptually complex. It is therefore implausible that the comprehension of speaker meaning should be driven by conscious reasoning, which would intolerably slow down the interpretation process. Human beings must, at some subpersonal level, be geared towards recognizing the inferential pathways (which we believe are largely metonymic) and apply them at “lightning speed” (Barcelona 2003). Such metonymic pathways are part of the cognitive competence of normal speakers and hearers and are readily accessible in particular linguistic and extra-linguistic contexts.[2] Given the largely unconscious nature of pragmatic inferencing, it does not make much sense to draw a clear-cut distinction between inferencing, on the one hand, and what is called spreading of activation, on the other, unless one wants to reserve the term ‘inferencing’ exclusively for deliberate conscious reasoning.[3]

One may ask at this point why there should be any inferentially based meaning at all. One plausible answer is that the “bottleneck” problem in linguistic communication has to be solved: The phonetic articulators are relatively slow in encoding information—about 7 syllables or 18 segments per second—conveying less than 100 bits per second of information (Levinson 2000: 6-7, 28). Stephen Levinson proposes that, from the speaker’s perspective, the solution to the bottleneck problem is to encode only the strict (but sufficient) minimum of information and leave the recovery of the full richness of intended meaning (including “default” meaning) to inferencing abilities of the hearer.[4] To quote Levinson (2000: 29): “[…] inference is cheap, articulation expensive, and thus the design requirements are for a system that maximizes inference.”

A characteristic of modern pragmatic approaches is that they try to account for inferential meanings on the basis of a restricted set of maxims or principles. For example, Levinson (1995, 2000) assumes that there are three abstract heuristics—quantity (Q), informativeness (I), and manner (M)—that guide the hearer in figuring out default meanings (generalized conversational implicatures). Horn (1984) reduces the Gricean maxims to two: Quantity (‘say as much as you can’) and Relevance (‘do not say more than you must’); and relevance theorists (e.g. Sperber & Wilson 1995, 2002) propose a single principle of relevance that is assumed to guide the hearers sufficiently in their efforts in “mind-reading” the intended meaning of speakers. Although the development of general principles that guide inferencing is certainly a desirable goal, there is a price to pay: When it comes to describing individual data, especially relevance theorists tend to resort to very detailed descriptions of how the pragmatic meaning of individual examples comes about—thus, in a way, belying their own highly abstract and generic principles of utterance interpretation.

Among the few generative linguists who have concerned themselves with conceptual structure, Jackendoff (1991, 2002) is, to our knowledge, the only one who integrates inferential rules into the semantic (conceptual) component. Jackendoff recognizes the need for postulating more concrete, metonymically based inferential processes, although he does not use the term ‘metonymy’ for these inferential principles. Consider, for example, how Jackendoff (2002: 387ff.) analyzes the well-known “ham sandwich” metonymy:

Jackendoff proposes an analysis of “enriched composition” for (1) with the reading ‘The person over in the corner contextually associated with a ham sandwich wants more coffee’ (388). Figure 1 provides a more formal representation.

Jackendoff is on the right track in assuming that the identification of the intended referent of the definite description in (1) involves a metonymic inference. But the inferential principle he proposes—object for person associated with object—overgenerates, leading to many unlikely or even infelicitous expressions. For example, the referents of the the pencil, the breadcrumb, or the lap top can all be objects “associated” with people in some way, but it is unlikely that they are used as metonymic expressions for referring to persons.

We argue in this article that abstract inferential principles based on Gricean maxims or, even more radically, on a unique principle of relevance cannot adequately account for how interlocutors actually proceed in inferring utterance meanings: There must be an intermediate level of inferential principles that are, on the one hand, abstract enough to be used as inferential schemata and, on the other hand, have enough specific conceptual content to serve as guideposts in inferential utterance interpretation. We advance the hypothesis that conceptual metonymies such as part-whole, cause-effect, person-role, representation-represented, which have been dubbed vital relations by Fauconnier and Turner (2002: 93ff.), are concrete enough to serve as reasoning principles in utterance interpretation. We regard such metonymic relations as multipurpose conceptual devices not restricted to language but used in other semiotic systems and thought as well.[5]

2. The basic metonymic relation[6]

Metonymy is often characterized as a ‘stand for’ relation (see e.g. Lakoff & Johnson 1980), a reflection of which is that metonymies are usually represented by the schema x for y, where x represents the source meaning (also called ‘vehicle’) and y symbolizes the target meaning of the metonymic operation. This “substitution” view of metonymy leads easily to the (erroneous) assumption that metonymy and pragmatic implicature are very different phenomena. An implicature is usually regarded as content that is added to what is said/explicitly conveyed. For example, in many contexts an expression such as widespread belief might trigger the implicature that the content of the belief is dubious, as in example (2):

(2) It is a widespread belief that linguists speak many languages +> ‘Linguists often do not speak many languages’

But things are not so clear. Levinson (2000) argues for the existence of a “heuristic” (similar to a Gricean maxim) “What is expressed simply is stereotypically exemplified”, which accounts for I-implicatures such as in (3)-(5) (adapted from Levinson 2000: 37):

(3) John’s book is good. +> ‘The book John read, wrote, borrowed, … is good’

Depending on one’s perspective, one could argue—in accordance with traditional conceptions of metonymy—that e.g. in (4) the meaning of female secretary is substituted for the source meaning of secretary; but one could also maintain that the meaning ‘female’ is added as a conceptual modifier to the meaning of secretary. We argue below that the crucial criterion for metonymy is not ‘addition’ or ‘substitution’ but the degree of conceptual prominence of the target meaning.

There is however also a tradition in linguistics that equates implicature with metonymically induced implication or that regards metonymy as a subtype of implicature. For example, in their introductory textbook to grammaticalization theory, Hopper and Traugott (1993) dedicate a whole chapter (ch. 4) to the significance of pragmatic inferencing, including metaphorical and metonymic inferencing, in the emergence of grammatical meanings out of lexical meanings.

In recent work it has been claimed that metonymy should not be viewed as a mere substitution relation. The research in Lakoff (1987), Radden and Kövecses (1999), Panther and Radden (1999), Langacker (2000), Barcelona (2000), Dirven and Pörings (2002) and Panther and Thornburg (2003) has shown that metonymy is better understood as a “reference point” (a vehicle or source) that triggers a target meaning. Francisco Ruiz de Mendoza and his colleagues at the University of La Rioja regard metonymy as a process of meaning elaboration that involves either expansion or reduction of a cognitive domain (matrix). This work emphasizes the conceptual nature of metonymy and is indeed an important step forward from the simplistic view of metonymy as a mere rhetorical trope to the insight that metonymy is a ubiquitous mental operation.

Simplifying somewhat the views expressed by these various authors, the basic metonymic relation can be diagrammed as in Figure 2:

In a linguistically manifest metonymic relation, a source meaning is related to a target meaning by means of a linguistic form (e.g. morpheme, word, phrase, sentence) that we call the linguistic vehicle.[7] The ellipse in Figure 2 represents the generally accepted assumption that the metonymic mapping takes place within one cognitive domain (ICM). Figure 2 also indicates that the source meaning is not obliterated by the target meaning, but still conceptually present (“salient”) or activated. Figure 2 does not indicate how stable or conventional the target meaning is. Indeed, the target meaning can be just a nonce sense, created on the spot, but it can also, through frequency of use, become a conventionalized meaning, stored separately in the mental lexicon. The result of this conventionalization of a metonymic target meaning is of course polysemy.

The noun phrase The Pentagon has two senses that are metonymically linked. On the one hand, there is a signifier-signified relation between the form and the source meaning ‘the pentagonal building housing the Defense Department of the U.S.’ and, on the other hand, as a result of the conventionalization of the target meaning ‘spokesperson for the Defense Department’, a signifier-signified relation between the same phonological form and the target meaning. This configuration is diagrammed in Figure 3 by two solid lines linking the same linguistic form with its two respective senses.

3. The contingent nature of the metonymic relation

Above we referred to the idea that conceptual metonymy is a “reference-point” phenomenon (Langacker 1993, 2000) where one conceptual entity provides access to another conceptual entity. This characterization—useful as it is—unfortunately overgeneralizes, i.e., it covers cases that in our view should not be treated as cases of metonymy. Sentences (7) and (8) illustrate the problem:

In sentence (7) the subject noun phrase the piano has the standard metonymic interpretation ‘the musician playing the piano’, with the meaning of piano providing mental access to the concept of piano player. Analogously, one could claim that in sentence (8), the sense of the loss of my wallet provides access to the concept of ‘non-possession (of the wallet)’. Are we therefore entitled to conclude that the relation between the concept of loss and that of non-possession is a metonymic relationship, just as the relation between the concept of piano and that of piano player is metonymic? Intuitively, the answer seems ‘no’; and in fact, there is an important difference between the two cases. In sentence (8) the relationship between ‘loss’ and ‘non-possession’ is conceptually necessary, i.e., the proposition presupposed by the referring expression the loss of my wallet in (8), ‘I lost my wallet at time t’, entails ‘I did not have my wallet for some time period beginning at time t’. In contrast, in sentence (7), the relationship between the piano and the piano player is contingent; the presupposition ‘There is a (contextually unique) piano’ does not entail ‘There is a piano player’. In other words, there is no metonymy loss for non-possession, but there is an often exploited metonymy musical instrument for musician.

The property of contingence that we claim characterizes metonymy is reminiscent of the property of defeasibility or cancelability of two well-known pragmatic implications, explicature and implicature. ‘Defeasibility’ and ‘contingence’ are however not necessarily synonymous: a relation between concepts may be contingent, i.e. conceptually non-necessary, but in a given linguistic and/or communicative context the target meaning may still be uncancelable. This is evident in sentence (7) where the meaning ‘piano player’ for piano does not seem to be defeasible in the given context.

There are other examples of non-cancelable metonymies—in particular, cases in which (i) grammatical construction meaning coerces lexical meaning and (ii) conversely, where lexical meaning coerces construction meaning. These types are illustrated in sections 3.1 and 3.2 below, respectively.

3.1 Constructionally coerced metonymies

We will now demonstrate how the meaning of a grammatical construction can coerce, i.e. enforce, a metonymic interpretation of a lexical expression. The relevant construction types are what we call action constructions, such as the imperative, and the How about VP and What about VP-ing constructions. These constructions (see Panther & Thornburg 2000, Ruiz de Mendoza & Pérez Hernández 2001) usually require an action predicate as in (9) and (10):

However, there are also naturally sounding utterances like (11) that contain a stative predicate:

One should normally not expect to find a stative predicate like be wealthy in an imperative construction. Nevertheless, sentence (11) receives an action interpretation, which can roughly be paraphrased as ‘Do something to the effect so that you will be wealthy in ten months’. The imperative construction seems to be responsible for the reinterpretation of the stative predication as an action. This phenomenon is known as coercion (Pustejovsky 1993) or semantic shift, which Leonard Talmy (2000: 324) defines as follows:

When the specifications of two forms in a sentence are in conflict, one kind of reconciliation is for the specification of one of the forms to change so as to come into accord with the other form.

In sentence (11) it is the specification (meaning) of (be) wealthy that changes to accord with the specification (meaning) of the imperative construction. This situation is diagrammed in Figure 4.

(Not represented in Figure 4 is the condition that a state can be viewed as being brought about intentionally.)

In cases like (11), the action interpretation is enforced, which seems to undermine our contention that metonymy is a contingent, i.e. in principle, defeasible, relation. But a closer look at sentence (11) reveals that the relation between a state and the action leading to that state is not conceptually necessary—any number of actions can lead to the same resultant state. One cannot—strictly speaking—logically infer “backwards” from a state to processes or actions that result in the state. That is, there is not a relation of entailment between ‘x is a state’ and ‘y is the action that leads to state x’. But one can make “reasonable” guesses; for example, in the case of Be wealthy in ten months one can think of a variety of actions (audacious investments, purchase of shares, etc.) that might lead to the desired result of being wealthy.

Thus, the relation between source and target remains contingent—it is in principle, but is not always de facto—defeasible. The context may however constitute an efficient barrier to cancellation. Now this is of course also a property that applies to implicatures: the context may enforce/coerce certain implicated interpretations.

3.2 Lexically coerced metonymies

One might be tempted to think that metonymic coercion always goes from grammatical meaning to lexical meaning as diagrammed in Figure 4. It would be nice if one could establish such unidirectionality of the coercion process. The notion of unidirectionality seems to underlie construction grammar, where it is assumed that constructions have meaning and that lexical items that are inserted in a construction do not necessarily have to fit “perfectly” but can, under certain circumstances, be coerced into a meaning determined by the construction meaning. But it is not impossible to imagine that lexical meaning might also “nibble at” constructional meaning and change it metonymically. To see this, consider (12):

One reading of (12) has the force of a directive speech act with an ‘action’ interpretation such as ‘Do something to the effect so that (as a result) you enjoy your summer vacation’. There might be folk models of the concept enjoyment that regard it as an experiential state that can be intentionally caused. Such an interpretation would be completely analogous to (11), involving the result for action metonymy. But there is also a folk model that does not regard enjoyment as a state that can be intentionally brought about, but rather as something that one experiences spontaneously. On the basis of this folk model, (12) would receive an optative interpretation with the meaning ‘S (=speaker) expresses the hope/wish that H (=hearer) will enjoy her summer vacation’. In this situation, the meaning attributed to enjoy—‘spontaneously occurring experiential state’—leads to a shift in construction meaning: the speech act component ‘H will do A’ (‘future action’ meaning) is discarded because it is incompatible with the mental state meaning of enjoy; only the compatible ‘wish/hope’ meaning remains, i.e., ‘S hopes that H will be in mental state s (enjoyment)’. Note that ‘wish’ is also a speech act component of prototypical imperatives, in which the speaker’s wish is directed towards a future action of the hearer. Since the action component has to be discarded, only the speaker’s wish that some future state-of-affairs obtains remains. The metonymic and coercive processes involved in (12) are set out in Figure 5.

An alternative approach to sentence (12)—more in line with construction grammar (Goldberg 1995)—would assume that the imperative construction itself is polysemous and that the optative interpretation of (12) is inherent in the construction; in this view enjoy simply fits the constructional meaning and can readily be inserted. Such an approach has the advantage of not having to abandon the hypothesis that coercion works from constructional meaning to lexical meaning, i.e. unidirectionally, but it has the disadvantage of proliferating polysemy in construction meanings.

Be that as it may, the main point with regard to our topic here is that the metonymic relation between the speech act concept directive and the mental concept wish is not conceptually necessary, but contingent, i.e., to repeat, in principle defeasible.

4. Pragmatic types of metonymy: referential, predicational, and illocutionary

Now we would like to turn to the question of how many types of metonymy there are. We view the question from a pragmatic angle. The starting-point is the often-heard claim that metonymies are typically a phenomenon of referential shift, i.e., in speech act terms, they are intimately tied to the referential act. (Examples (9)-(12), discussed above, do not fall into the category of referential metonymies, by the way.) We have already seen an example of typical referential metonymy in sentence (6) where the metonymy place for institution helps to identify the intended referent of the Pentagon. Sentence (13) is another example:

In (13) the company name General Motors is used to refer to the automobile workers who walk out of the work place.

One can however find metonymies in other than referential functions. Here we will briefly mention two additional pragmatic types, predicational metonymy and speech act or illocutionary metonymy, and argue for treating them as genuine metonymies.[8] An example of a predicational metonymy is:

In (14) the necessity or obligation to stop production evokes the actual occurrence of the event of stopping production (obligation to act for action). The inference involved is an instance of a high-level metonymic principle that is very common in English and other languages especially when the modality is in the past: A potential event (e.g. the ability, possibility, permission, obligation to undertake an action) is metonymically linked to its occurrence in reality. Events are conceptualized here as idealized cognitive models (ICMs) that contain as subcomponents the modalities of their realization. Sentence (14) also illustrates a propositional metonymy because both the referring expression General Motors (‘the executive officers of GM’) and the predicating expression had to stop production (‘stopped production’) undergo a metonymic shift. Note again that these shifts in reference and predication are not conceptually necessary but contingent (i.e. in principle, cancelable).

Finally, we also assume the existence of illocutionary metonymies. The well-known phenomenon of indirect speech acts can be accounted for on a metonymic basis:

In utterance (15) the expression of the wish of the speaker with regard to the action to be carried out by the addressee (signaled by would like you to) metonymically evokes the request to close the window itself (see Gibbs 1994, 1999; Thornburg & Panther 1997, Panther & Thornburg 1998, 2003b; and Ruiz de Mendoza & Pérez Hernández 2001, 2003). The basic idea is that an attribute of a speech act can stand for the speech act itself in the same way that an attribute of a person can stand for the person. Figure 6 provides a schematized representation of how utterances of type (15) might activate the illocutionary force of a directive, e.g. a request. Note that this example shows that propositional forms can be linked metonymically.

Still, one might doubt that what we call referential metonymies, predicational metonymies, and illocutionary metonymies are really of the same type. Our contention that the relations in (13) between General Motors and ‘the workers employed by General Motors’, on the one hand, and that in (14) between had to stop production and ‘(actually) stopped production’, on the other hand, are of the same kind, viz. metonymic, may seem surprising. One might object that the target meaning of (14) is “really” an implicature that comes about through pragmatic strengthening of the proposition expressed in it.

Our answer to this objection is: First, a metonymic analysis does not preclude a pragmatic analysis in terms of conversational implicature. On the contrary, we assume that conversational implicatures, or more generally, pragmatic inferences, are often guided by preexisting metonymic principles.[9] Second, the same metonymy can be triggered predicationally and referentially.[10] For example, the obligation to act for action metonymy is not only operative on predicational vehicles but can also be triggered by the nominalized (referential) counterpart of the predicate in (14), viz. the italicized noun phrase in (16):

(16) General Motor’s obligation to stop production had a devastating effect on the economy.

Utterance (16) very strongly suggests that General Motors actually did stop production. The target meaning of the referring expression in (16) can thus be paraphrased as ‘the fact that General Motors stopped production’. And it seems that the predicate had a devastating effect on the economy is interpreted as the consequence of the actual stopping of production, rather than just of the obligation to stop it.

Third, even illocutionary metonymies find their analogues in referential positions. Sentence (17) – I am willing to lend you my car – , which may trigger the target meaning ‘I offer to lend you my car’, is paralleled by a referential metonymy triggered by the nominalized expression in sentence (18):

The referential noun phrase in (18) lends itself quite readily to the (defeasible) target meaning ‘my offer to lend you my car’. Thus, there does not seem to be any reason to treat the inference that can be drawn from the content of the referential subject noun phrase differently from the target meaning of well known uncontroversial metonymies as in utterances like Table Four wants another Chardonnay, where Table Four stands for ‘the customer sitting at Table Four’.

5. Conceptual prominence of the metonymic target

As we pointed out earlier, the traditional definition of metonymy as a substitution relation has been rightly criticized by cognitive linguists (cf. Radden & Kövecses 1999) and instead a view of metonymy as a reference-point phenomenon has been suggested, which is a step forward, but has its own problems in being too general. Our view is that typical metonymies involve what we call conceptual prominence of the target. To see how this works, consider utterance (19):

(19) General Motors had to stop production on Monday but they resumed it on Thursday.

The but-clause in (19) makes pragmatic sense only if the clause General Motors had to stop production on Monday has the prominent metonymically derived reading ‘General Motors stopped production on Monday’. The source meaning of the first clause in (19) (the ‘obligation’ sense) is certainly active, but the relevant sense is the target meaning, because it is only against the ‘factuality’ sense of the first clause that the second clause can be interpreted in a reasonable way.

Also consider sentence (20) from a newspaper article, whose metonymic structure is sketched in Figure 7:

(20) North Korea’s willingness to publicly flout its international commitments suggests it is trying to force itself onto Washington’s agenda to win an oft-stated goal: talks with its longtime foe about a nonaggression treaty. [The Southern Illinoisan, 26 December 2002]

From the context it is clear that the writer of example (20) intends to convey the idea that North Korea is not only willing to flout its international commitments, but that it actually does flout them. In other words, we have a highly productive metonymically induced inferential principle here: willingness to act for (actual) action. Moreover, despite the high degree of activation or salience of the source meaning, the target meaning seems to be conceptually more important and relevant than the source meaning. What the whole newspaper article is about is not so much what North Korea is willing to do as to what it has already done and will do in terms of nuclear weapons development.

To summarize, we contend that in a prototypical metonymy the target meaning is more prominent than the source meaning, although the source meaning must of course have a sufficient degree of salience in the context of the utterance in order to be able to activate the target. But the raison d’être of metonymy is to make the target not only accessible, as suggested by the reference-point theory of metonymy, but, just as importantly, to make it available for the ensuing discourse. As can be seen in example (20) above, the assumption (metonymic target) that North Korea has already developed or will develop the nuclear weapons is the starting-point of future debates about what can be done about this dangerous situation.

If it is the case that the relatively greater conceptual prominence of the target meaning is a feature of prototypical metonymies, the traditional view of metonymy as a ‘stand-for’, i.e. a substitution relation, does not look so wrong after all. In this perspective, substitution of the target for the source meaning is the borderline case where the target meaning has become maximally prominent. When this happens, there is no metonymic relation anymore, because the source meaning has simply been supplanted by the target meaning.

The property of conceptual prominence postulated here for prototypical conceptual metonymies seems to be related to what Erteschik-Shir (1979: 443)) calls dominance (of a syntactic constituent) in a different context. A constituent is called dominant in an utterance if and only if the speaker intends to direct the attention of the hearer to the conceptual content of the constituent. The dominant constituent becomes “the natural candidate for the topic of further conversation” (443). A procedure for testing dominance is the reaction of a speaker B to the sentence uttered by a speaker A. B responds by a sentence in which the dominant constituent X is assigned a truth, a probability, or an interest value as in example (21):

Speaker B’s reaction to speaker A’s utterance may relate to the truth-value of the embedded proposition that Mary kissed Bill; in other words, the complement clause is the object of speaker B’s judgment. (Of course, B’s utterance may also be a truth evaluation of the proposition expressed by A as a whole). In the metonymic framework adopted here, B’s reaction is based on an interpretation of A’s utterance guided by the metonymy attributed assertion for assertion, i.e. the proposition asserted by John is treated as if it had been asserted by speaker A. The interesting point about such examples as (21) is that a metonymically implied concept is conceptually as prominent as, or even more prominent than, its explicitly expressed source concept.

In light of what we have said about the conceptual prominence of the target in prototypical metonymies, it seems that some cases that have often been adduced as typical examples of metonymy are not such good examples after all. Consider the hoary example

which is usually analyzed as exemplifying the metonymy controller for controlled. As Lakoff and Johnson (1980: 39) point out, Nixon himself did not drop the bombs on Hanoi, but he was ultimately responsible for this military action. In other words, the referent designated by the source meaning is the ultimate causer of the action. However, it is not the rather indeterminate target meaning that is conceptually prominent, but the source meaning itself (see Figure 8):

In (23) what is conceptually prominent is the target meaning, not the source meaning. Sentence (23) is about the saxophone player, not about the saxophone. In contrast, sentence (22) is really about Nixon, and not about the pilots that bomb Hanoi. This intuition is confirmed by coreference facts. It is quite natural (and cynical) to say:

(24) In the morning, Nixon bombed Hanoi; at noon he (= Nixon) had lunch with aides. (Topic: Nixon himself)

In contrast, (25) where they is supposed to refer to the target is rather odd:

(25) ?#In the morning, Nixon bombed Hanoi; at noon they (= the pilots) were on some other mission.

The situation is reversed in the case where the target meaning is conceptually prominent:

(26) The sax has the flu today and he (= the saxophone player) will not be able to play tonight. (Topic: the saxophone player)

(27) ?#The sax has the flu today but it (= the instrument) needs repair anyway.

In (26) he in the second clause refers to the target of the sax in the first clause; there is topic continuity. The whole sentence is about the saxophone player, not the saxophone. Sentence (27) is however rather disruptive because in the first clause the target ‘the saxophone player’ is talked about but in the second clause there is a sudden referential shift to the instrument.

We conclude that the object used for user (or more specifically: instrument for musician) metonymy is a prototypical metonymy because it makes the target conceptually more prominent than the source whereas the ultimate causer for immediate causer is a more peripheral metonymic relation because the source is conceptually more prominent than the target.

Ruiz de Mendoza and Díez Velasco (forthcoming) explain these coreference phenomena in terms of the relative scope of cognitive domains (source and target). Their Domain Availability Principle postulates that it is always the matrix domain, i.e. the most-inclusive domain that determines the properties of metonymic anaphoric reference. In their terminology, in (24) the source domain that Nixon constitutes is a larger domain than his air force and the anaphor to be used is therefore he. In (26), the target domain ‘the saxophone player’ is assumed to be more inclusive than the source domain that is literally designated by the sax, and again the most natural pronoun choice is he (or she as the case may be). Note that in both cases the matrix domain is human and it seems to us that this is the reason why, in one case, the source domain is more inclusive and in the other case, the target domain is. Humans tend to be associated with the larger cognitive domains and everything else tends to be defined in relation to humans.

What we have said so far about conceptual prominence, coreference, and topicality may also shed light on the problem of identifying the locus of a conceptual metonymy:[11]

The first possibility is that (28) is a predicational metonymy where the manner of speaking (brief) stands for the speech event itself (see Figure 10):

However, there is also the possibility that the subject term is metonymically interpreted, i.e. that (28) exemplifies a referential metonymy, as diagrammed in Figure 11:

The reading of (28) would thus be that the speech given by the president was brief.

Now, is there any way of deciding between these two competing analyses? We think there is. In English, the evidence speaks for an analysis in terms of Figure 10, i.e. for a predicational metonymy. To see this, let us first test the ‘referential metonymy’ hypothesis, i.e. assume that the metonymic target meaning of president is ‘president’s speech’ in the first clause of (29) and (30) respectively.[12] Now consider the following coreference facts:

(29) #The president_s®t was brief and Ø_tdid not contain any interesting thoughts.

Intended reading: ‘The president’s speech was brief and did not contain any interesting thoughts’

(30) #The president_s®t was brief but it_tcontained a number of interesting thoughts.

Intended reading: ‘The president’s speech was brief but it contained a number of interesting thoughts’

If the president in (29) and (30) has the metonymic reading ‘the president’s speech’, one would expect the zero anaphor in (29) and the pronoun in (30) to be coreferential with the president’s speech. However, there is clearly a break in coherence in both (29) and (30) between the first clause (interpreted as referring to the target ‘president’s speech’) and the second clause where something is said about that target. This seriously undermines the interpretation of (28) as a case of (prototypical) referential metonymy, at least if one assumes that the target is the prominent conceptual entity—in accordance with our definition of prototypical metonymy—and the topic in the ensuing discourse.

Let us now examine the possibility that the metonymy in (28) is predicational, as diagrammed in Figure 10 (same subscripts denote referential identity as before):

(31) The president_i was brief_s_®t about the issue of tax cuts and Ø_ileft the meeting.

‘The president spoke briefly about the issue of tax cuts and left the meeting’

(32) The president_i was brief_s_®t about the issue of tax cuts because he_ihad a lunch appointment.

‘The president spoke briefly about the issue of tax cuts because he had a lunch appointment’

In this case, by hypothesis, we assume that there is no referential shift from the president to ‘the president’s speech’ but that the metonymic shift occurs in the predicate: i.e., was brief about NP is metonymically interpreted as ‘spoke briefly about NP’. Both (31) and (32) are completely natural with a non-metonymic interpretation of the president. We conclude that in a sentence of the type The president was brief about NP, the human referent of the subject—the president—is not metonymic and naturally determines the anaphoric structure of sentences (29)–(32).[13]

The predicational metonymy analysis is further supported by the fact that brief can be modified by a manner adverb denoting ‘intention’. An adverb such as deliberately normally modifies an action verb that requires a rational agent as an argument. This fact points to a ‘linguistic action’ reading of the metonymic vehicle brief, viz. ‘speak briefly’, as in (33):

(33) The president was deliberately brief about the issue of tax cuts because he had a lunch appointment.

Finally, there are coordination facts that speak in favor of a predicational metonymy analysis for (28), as in (34):

(34) The chancellor was [brief about the tax cut] but [spoke for hours about health reform].

Normally only constituents of the same syntactic and preferably the same semantic type are coordinated. If it is assumed that the target of brief is an action concept, then the syntactic and semantic requirements for coordinating the two verb phrases in (34) are satisfied.

To summarize, considerations of topicality, coreference restrictions and coordination constraints support an analysis where the adjectival predicate in (28) is selected as the locus of metonymic elaboration. What is conceptually prominent here is the brevity of the president’s speech but this target meaning is triggered by the predicate was brief about NP rather than the referential subject the president.

6. Conclusion: Properties of prototypical conceptual metonymy

We hope to have made a plausible case for the idea that conceptual metonymies constitute an intermediate level of contingent conceptual relations—between very abstract inference-guiding principles and heuristics à la Sperber and Wilson and Levinson (and perhaps very specific ad hoc inferential principles that are employed in the derivation of particularized conversational implicatures). Many examples we have analyzed as metonymies in this article are canonically regarded as explicatures or implicatures in the pragmatic literature. We have no objection to such an analysis but have argued that such pragmatic inferences are often guided by pre-existing conceptual metonymies readily available to interlocutors in their interpretive efforts. Cross-linguistic comparisons, which we have not undertaken in this paper, seem to indicate that the degree of exploitation of metonymic principles may vary from language to language (see e.g. Panther & Thornburg 1999, 2000; Brdar & Brdar-Szabó 2003).

Furthermore, we have developed the idea that in prototypical metonymic relations the target concept is conceptually prominent. Prototypical metonymy not only makes target meanings accessible but also available for further elaboration in discourse. Metonymies function on the referential, predicational and illocutionary levels of speech acts. They also perform an important function in resolving semantic conflicts between lexical meaning and constructional meaning. Assuming the existence of a separate layer of metonymic inferencing has one obvious advantage: It provides hearers with sufficiently concrete inferential pathways, i.e. natural inference schemas, not derivable in a straightforward fashion from either Gricean maxims or the principle of relevance. Lastly, we have demonstrated that conceptual metonymy is locatable in both conventional(ized) meaning and is used on the fly in the construction of utterance meaning. It thus cuts across the traditional distinction between semantics and pragmatics.

7. References

Barcelona, Antonio (ed.) (2000): Metaphor and Metonymy at the Crossroads: A Cognitive Perspective, Berlin and New York.

Barcelona, Antonio: (2003): “The case for a metonymic basis of pragmatic inferencing: evidence from jokes and funny anecdotes”, in: Panther, Klaus-Uwe / Thornburg, Linda (edd): Metonymy and Pragmatic Inferencing, Amsterdam and Philadelphia, 81-102.

Brdar, Mario / Brdar-Szabó, Rita (2003): “Metonymic coding of linguistic action in English, Croatian and Hungarian”, in: Panther, Klaus-Uwe / Thornburg, Linda L. (2003): Metonymy and Pragmatic Inferencing, Amsterdam and Philadelphia, 241-266.

Chybowska, Urszula (2004): “The role of the hearer in the metonymic structure of the directive scenario.” Talk given at the Warsaw-Hamburg Cognitive Linguistics Workshop. Warsaw, Poland, January 16-17. (urs_zula@poczta.onet.pl)

Dirven, René / Pörings, Ralf (edd.) (2002): Metaphor and Metonymy in Comparison and Contrast, Berlin and New York.

Erteschik-Shir, Naomi (1979): “Discourse constraints on dative movement”, in: T. Givón (ed.): Discourse and Syntax, New York, 441-467.

Fauconnier, Gilles / Turner, Mark (2002): The Way We Think: Conceptual Blending and the Mind’s Hidden Complexities, New York.

Gibbs, Raymond W., Jr. (1994): The Poetics of Mind: Figurative Thought, Language, and Understanding, Cambridge.

Gibbs, Raymond W., Jr. 1999. “Speaking and thinking with metonymy”, in: Klaus-Uwe / Thornburg, Linda (edd): Metonymy and Pragmatic Inferencing, Amsterdam and Philadelphia, 61-76.

Goldberg, Adele (1995): Constructions: A Construction Grammar Approach to Argument Structure, Chicago and London.

Horn, Laurence R. (1984): “Toward a new taxonomy for pragmatic inference: Q- and R-based implicature”, in: Schiffrin, D. (ed.): Meaning, Form, and Use in Context, Washington DC, 11-42.

Jackendoff, Ray (2002): Foundations of Language: Brain, Meaning, Grammar, Evolution, Oxford.

Lakoff, George (1987): Women, Fire, and Dangerous Things: What Categories Reveal About the Mind, Chicago and London.

Lakoff, George / Johnson, Mark (1980): Metaphors We Live By, Chicago and London.

Langacker, Ronald W. (1993): “Reference-point constructions”, in: Cognitive Linguistics 4, 1-38.

Langacker, Ronald W. (2000): Grammar and Conceptualization, Berlin and New York.

Levinson, Stephen C. (1995): “Three levels of meaning”, in: Palmer, F. (ed.): Grammar and Meaning, Cambridge, 90-115.

Levinson, Stephen C. (2000): Presumptive Meanings: The Theory of Generalized Conversational Implicature, Cambridge, MA and London.

Panther, Klaus-Uwe / Radden, Günter (edd) (1999): Metonymy in Language and Thought, Amsterdam and Philadelphia.

Panther, Klaus-Uwe / Thornburg, Linda L. (1998): “A cognitive approach to inferencing in conversation”, in: Journal of Pragmatics 30, 755-769.

Panther, Klaus-Uwe / Thornburg, Linda L. (1999): “Coercion and metonymy: The interaction of constructional and lexical meaning”, in: Lewandowska-Tomaszczyk, B. (ed.): Cognitive Perspectives on Language, Frankfurt am Main, 37-52

Panther, Klaus-Uwe / Thornburg, Linda L. (2000): “The effect for cause metonymy in English grammar”, in: Barcelona, A. (ed.): Metaphor and Metonymy at the Crossroads: A Cognitive Perspective, Berlin and New York, 215-231.

Panther, Klaus-Uwe / Thornburg, Linda L. (edd). (2003): Metonymy and Pragmatic Inferencing, Amsterdam and Philadelphia.

Panther, Klaus-Uwe / Thornburg, Linda L. (2003a): “Introduction: On the nature of conceptual metonymy”, in: Panther, Klaus-Uwe / Thornburg, Linda L. (edd).: Metonymy and Pragmatic Inferencing, Amsterdam and Philadelphia, 1-20.

Panther, Klaus-Uwe / Thornburg, Linda L. (2003b): “Metonymies as natural inference and activation schemas: The case of dependent clauses as independent speech acts”, in: Panther, Klaus-Uwe / Thornburg, Linda L. (edd). (2003): Metonymy and Pragmatic Inferencing, Amsterdam and Philadelphia, 127-147.

Panther, Klaus-Uwe / Thornburg, Linda L. (Forthcoming): “Metonymy”, in: Geeraerts, D. / Cuyckens, H. (edd): Handbook of Cognitive Linguistics [Chapter 10], Oxford.

Pustejovsky, James (1993): “Type coercion and lexical selection”, in: Pustejovsky, J. (ed.): Semantics and the Lexicon, Dordrecht, 73-96

Radden, Günter / Kövecses, Zoltán (1999): “Towards a theory of metonymy”, in: Panther, Klaus-Uwe / Radden, Günter (edd): Metonymy in Language and Thought, Amsterdam and Philadelphia, 17-59.

Récanati, François (2002): “Does linguistic communication rest on inference?”, in: Mind and Language 17, 105-126.

Ruiz de Mendoza, Francisco J. / Pérez Hernández, Lorena (2001): “Metonymy and the grammar: motivation, constraints and interaction”, in: Language and Communication 21, 321-357.

Ruiz de Mendoza, Francisco J. / Pérez Hernández, Lorena (2003): “Cognitive operations and pragmatic implication”, in: Panther, Klaus-Uwe / Thornburg, Linda L. (edd): Metonymy and Pragmatic Inferencing, Amsterdam and Philadelphia, 23-49.

Ruiz de Mendoza, Francisco J. / Otal Campo, José L. (2002): Metonymy, Grammar, and Communication, Albolote and Granada.

Ruiz de Mendoza, Francisco J. / Díez Velasco, Olga (Forthcoming): “Metonymic types and anaphoric reference”, in: Radden, G. / Panther, K.-U. (edd.): Studies in Linguistic Motivation, Berlin and New York.

Sperber, Dan / Wilson, Deidre (1995): Relevance: Communication and Cognition, Oxford.

Sperber, Dan / Wilson, Deidre (2002): “Pragmatics, modularity and mind-reading”, in: Mind & Language 17, 3-23.

Talmy, Leonard (2000): Toward a Cognitive Semantics. Vol. II: Typology and Process in Concept Structuring, Cambridge, MA and London.

Thornburg, Linda / Panther Klaus-Uwe (1997): “Speech act metonymies”, in: Liebert, W.-A. / Redeker, G. / Waugh, L. (edd): Discourse and Perspectives in Cognitive Linguistics, Amsterdam and Philadelphia, 205-219.

[1] This paper is a revised version of talks delivered at the 8^th International Cognitive Linguistics Conference at the University of La Rioja, Logroño, Spain, July 20-25, 2003, and at the Warsaw-Hamburg Cognitive Linguistics Workshop, Warsaw, Poland, January 16-17, 2004.

[2] Our claim that pragmatic inferencing is usually automatic and subconscious does not of course preclude the possibility that there are communicative situations in which a hearer will resort to conscious reasoning in order to figure out what a speaker might mean on a specific occasion. See Chybowska (2004) for further analysis of the role of the hearer in indirect speech act interpretation.

[3] Récanati (2002) has hypothesized recently that “primary pragmatic processes” involved in explicature derivation are as “direct” as perception and that “secondary primary processes” (inferring implicatures) are amenable to conscious reasoning (Récanati’s availability condition).

[4] For Levinson, the key players in this process are generalized conversational implicatures (GCIs).

[5] Sperber and Wilson (2002) have however recently proposed that the kind of inferencing used in linguistic communication is modular, i.e. specific to language.

[6] The following is based on Panther and Thornburg (2003a) and Panther and Thornburg (forthcoming).

[7] Not all theories make a distinction between vehicle and source. For example, Radden and Kövecses (1999) use the term ‘vehicle’ to refer to the linguistic sign that triggers the metonymic relation as a whole. Ruiz de Mendoza and Otal (2002) and Barcelona (2000) use the term ‘source’ in the same function. Radden and Kövecses (1999) regard a number of other relations as metonymic including relationships of form as e.g. in euphemistic usages such as Gosh for ‘God’ or shoot for ‘shit’, which we regard as more peripheral examples of metonymy.

[8] See Searle (1969) for the relevant distinctions between referring and predicating—which together form the propositional act—and illocution.

[9] This is not a claim that metonymic principles are innate, but rather pre- and extra-linguistic. That metonymic principles guide the production and comprehension of pragmatic inferences is e.g. argued for by Ruiz de Mendoza and Pérez Hernández (2003) for explicature derivation and by Barcelona (2003) for more indirect pragmatic implications.

[10] There are constraints governing the parallelism between predicational and referential metonymies that, to our knowledge, have not been studied systematically.

[11] Brdrar and Brdar-Szabó (2003) raise the same question from a typological perspective in an interesting cross-linguistic study comparing the exploitation of the manner for linguistic action metonymy in Croatian, Hungarian and English.

[12] We use the subscripts s for ‘source’ and t for ‘target’ referents. The arrow ‘®’ indicates the metonymic relationship.

[13] In Langacker’s framework of cognitive grammar, brief would function as the (scalar) landmark in relation to a trajector that is a process (the president speaking), not a participant (the president) in the process.

The Role of Conceptual Metonymy in Meaning Construction [1]