Aspects of Meaning Construction (Z 136)

Aspects of Meaning Construction Aspects of Meaning Construction Edited by Günter Radden University of Hamburg Kla...

Author: Gunter Radden | Klaus-Michael Kopcke | Thomas Berg | Peter Siemund

41 downloads 1877 Views 2MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Aspects of Meaning Construction

Aspects of Meaning Construction Edited by

Günter Radden University of Hamburg

Klaus-Michael Köpcke University of Münster

Thomas Berg Peter Siemund University of Hamburg

John Benjamins Publishing Company Amsterdam/Philadelphia

8

TM

The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences – Permanence of Paper for Printed Library Materials, ansi z39.48-1984.

Library of Congress Cataloging-in-Publication Data Aspects of meaning construction / edited by Günter Radden, Klaus-Michael Köpcke, Thomas Berg and Peter Siemund. p. cm. Includes bibliographical references and indexes. 1. Semantics. 2. Metonyms. 3. Metaphor. 4. Grammar, Comparative and general. O. Radden, Günter. P325.A775 2007 401.43--dc22 isbn 978 90 272 3242 7 (Hb; alk. paper)

2006047946

© 2007 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microﬁlm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa

For Klaus-Uwe Panther on the occasion of his 65th birthday

Table of contents List of contributors Introduction: The construction of meaning in language

ix 1

part 1 Metonymy and metaphor 1. Experimental tests of figurative meaning construction Raymond Gibbs, Jr.

19

2. High-level metaphor and metonymy in meaning construction Francisco José Ruiz de Mendoza Ibáñez and Ricardo Mairal Usón

33

3. The role of metonymy in meaning construction at discourse level: A case study Antonio Barcelona

51

4. Chained metonymies in lexicon and grammar: A cross-linguistic perspective on body part terms Martin Hilpert

77

5. Arguing the case against coercion Debra Ziegeler 6. When Zidane is not simply Zidane, and Bill Gates is not just Bill Gates: Some thoughts on the construction of metaphtonymic meanings of proper names Mario Brdar and Rita Brdar-Szabó 7. Collocational overlap can guide metaphor interpretation Anatol Stefanowitsch

99

125 143

part 2 Mental spaces and conceptual blending 8. Constructing the meanings of personal pronouns Ronald W. Langacker 9. The construction of meaning in relative clauses: Indeterminacy and constraints Kiki Nikiforidou

171

189

viii Aspects of Meaning Construction

10. Constraints on inferential constructions Christian Koops

207

11. The construction of vagueness: “Sort of ” expressions in Romance languages Wiltrud Mihatsch

225

12. Communication or memory mismatch: Towards a cognitive typology of questions Wolfgang Schulze

247

13. Brutal Brits and persuasive Americans: Variety-specific meaning construction in the into-causative Stefanie Wulff, Anatol Stefanowitsch, and Stefan Th. Gries

265

Index of authors Index of subjects Index of metonymies and metaphors

283 285 289

List of contributors Antonio Barcelona Departamento de Filologia Inglesa Universidad de Murcia E-30071 Murcia Spain E-mail: [email protected] Thomas Berg University of Hamburg Institute of English and American Studies Von-Melle-Park 6 D-20146 Hamburg Germany E-mail: [email protected] Mario Brdar Department of English Language and Literature Jossip Juraj Strossmayer University Lorenzo Jägera 9 HR-31000 Osijek Croatia E-mail: [email protected] Rita Brdar-Szabó Germanistikai Intézet Eötvös Loránd University 19-21 Ajtósi Dürer sor H-1146 Budapest Hungary E-mail: [email protected] Raymond Gibbs Department of Psychology University of California, Santa Cruz Santa Cruz, CA 95064 USA E-mail: [email protected] Stefan Th. Gries Department of Linguistics University of California, Santa Barbara Santa Barbara, CA 93106-3100

USA E-mail: [email protected] Martin Hilpert Rice University Department of Linguistics MS 23 6100 Main Street Houston, TX 77005-1892 USA E-mail: [email protected] Christian Koops Rice University Department of Linguistics – MS 23 P.O. Box 1892 Houston, TX 77251-1892 USA E-mail: [email protected] Klaus-Michael Köpcke Westfälische Wilhelms-Universität Münster Germanistisches Institut Leonardo Campus 11 D-48149 Münster Germany E-mail: [email protected] Ronald W. Langacker University of California San Diego Linguistics, 0108 La Jolla, CA 92093-0108, USA E-mail: [email protected] Ricardo Mairal Usón Despacho 618 Departamento de Filologías Extranjeras y sus Lingüísticas Facultad de Filología Senda del Rey, 7 E-28040 Madrid Spain E-mail: [email protected]

Aspects of Meaning Construction Wiltrud Mihatsch Fakultät für Linguistik und Literaturwissenschaft University of Bielefeld Postfach 100131 D-33501 Bielefeld Germany E-mail: [email protected] Kiki Nikiforidou 24 Argirokastrou St., Vrilissia GR-Athens 152 35 Greece E-mail: [email protected] Günter Radden University of Hamburg Institute of English and American Studies Von-Melle-Park 6 D-20146 Hamburg Germany E-mail: [email protected] Francisco José Ruiz de Mendoza Faculty of Letters and Education University of La Rioja Edificio Quintiliano c/ La Cigüeña 60 Campus Universitario E-26004, Logroño, La Rioja Spain E-mail: [email protected] Wolfgang Schulze Institut für Allgemeine und Typologische Sprachwissenschaft Dept. II, F 13/14 Ludwig-Maximilians-Universität München Geschwister-Scholl-Platz 1 D-80539 München Germany E-mail: [email protected] Peter Siemund University of Hamburg Institute of English and American Studies Von-Melle-Park 6 D-20146 Hamburg Germany E-mail: [email protected]

Anatol Stefanowitsch University of Bremen Bibliothekstrasse / GW2 D-28334 Bremen Germany E-mail: [email protected] Stefanie Wulff Department of Linguistics University of California at Santa Barbara Santa Barbara, CA 93109-3100 USA E-mail: [email protected] Debra Ziegeler Suite 189, Lansdowne House 792 Wilmslow Road Manchester M20 6UG UK E-mail: [email protected]

introduction

The construction of meaning in language Günter Radden, Klaus-Michael Köpcke, Thomas Berg, Peter Siemund University of Hamburg/University of Münster/University of Hamburg

1. Underspecification and the construction of meaning It is unanimously agreed in Cognitive Linguistics that meaning does not reside in linguistic units but is constructed in the minds of the language users. For the listener this means that he takes linguistic units as prompts and constructs from them a meaningful conceptual representation. In fact, this principle of meaning construction is not confined to language. Every transformation of a sensory stimulus into a mental representation is an instance of meaning construction, which is rooted in the interaction of human beings with their environment. The world around us is not meaningful per se but rather acquires meaning through the human mind. For example, the moon is not in itself meaningful but it may be made purposeful in very different ways: astronomers and lovers have fairly disparate views of the moon. The fundamental principles by which the mental transformation of a sensory stimulus is effected were discovered by Gestalt psychologists in the early twentieth century. Sensory stimuli tend to be interpreted in such a way that we see “good” gestalts. Figure 1 illustrates the presumably universal Gestalt principle of closure. The pattern of equidistant dots in (a) is typically seen as a square in (b). That is, the lines connecting the dots are in the eyes of the beholder: they are mentally added to the visual input and transform it into a meaningful Gestalt. A holistic figure such as a square is simply more meaningful to us than four isolated dots. Gestalt principles are fundamental devices of meaning construction; they operate automatically and beyond our awareness and are probably shared by all human beings. We may compare the individual dots in (a) to a string of linguistic units and the

a. Equidistant dots

b. Square

Figure 1. Gestalt principle of closure


⇒ a. “Meaningless” figure

b. Two carpenter’s squares

c. Picture frame

d. Picture frame

Figure 2. Reproductions of a seemingly meaningless figure

square in (b) to the conceptual representation which is evoked by these linguistic units. Meaning construction pervades every aspect of our lives. A crucial aspect of our interaction in the world is being able to identify and categorize things. In his pioneering work on remembering, the British psychologist Sir Frederic Bartlett (1932) confronted subjects with what would seem to be meaningless figures and asked them to remember and reproduce them. One of the figures they were shown was Figure 2a. Some subjects would reproduce the figure the way it was presented to them, i.e. as (b), others would reproduce it as (c) or (d). The striking observation Bartlett made was that, in reproducing meaningless figures, his subjects turned them into meaningful drawings. For example, some subjects would interpret figure (a) as “two carpenter’s squares” and hence reproduce it faithfully as in (b); other subjects would interpret it as a “picture-frame” and reproduce it as in (c) or (d). Bartlett fittingly described people’s desire to associate things with meaning as “effort after meaning.” Interestingly, finding a name for a meaningless figure turned out to be very helpful in the subjects’ effort after meaning: they often felt relieved when they found a label which expressed a certain concept. In Langacker’s (1987) terminology, the subjects were able to project a “sanctioning structure,” i.e. the concept of two carpenter’s squares or a picture frame, to the meaningless figures. We reinterpret apparent meaninglessness in terms of underspecification: Bartlett’s subjects imposed an interpretation on the stimulus figures which are somehow underspecified. This can be seen from the fact that, within a certain range, subjects differ in their interpretation of what they see. This range is, amongst others, constrained by the figure itself. Underspecification is especially relevant for the interpretation of linguistic material. We believe that each utterance is underspecified in the sense that it cannot possibly express all aspects that are relevant to its full interpretation. Underspecification in language is not just the result of time constraints but is governed by communicative principles. For example, according to Grice’s maxim of quantity, an underspecified utterance means something different from a fully explicated utterance. In his iconic approach to text interpretation, Posner (1986: 305f) provides a convincing example of differing interpretations derived from underspecified and more explicitly specified descriptions of the same situation:

Introduction: The construction of meaning in language

(1) a. Mr. Smith stopped in front of his house. He waved to a passing neighbour and got out of his car. b. Mr. Smith stopped in front of his house. He raised his arm and smiled to a passing neighbour. He pulled the handle of the door of his car, pushed the door open, swung his legs out, heaved his body out, and shut the door.

The two passages differ with respect to the detail of description. Description (1a) is less detailed, but the reader will easily mentally supply the actions that are explicated in description (1b). Moreover, the reader understands Mr. Smith to be a normal adult and his actions as everyday behaviour. The painstaking description (1b), by contrast, is not just more detailed but suggests great effort involved in the actions. Hence Mr. Smith tends to be seen as a cautious elderly gentleman who plans every step carefully. None of these meanings are explicitly expressed in the description— they are inferred, or constructed. Linguistic units almost invariably underspecify the conceptualizations they code (Langacker 1987: 66). In his study on “The windowing of attention in language,” Talmy (1996) has shown that the gapping of conceptual material in event frames may even be grammatically required. For example, in causal events we normally window our attention on the ultimate result. In expressing such events, we give expression to the final event, possibly also express the penultimate subevent, but have to gap the subevents in the middle of the causal chain. Thus, I broke the window by hitting it with a rock represents the normal way of expressing a causal event, mentioning the penultimate event is of marginal acceptability (?I broke the window by throwing a rock at it), and mentioning an intermediate subevent makes the sentence unacceptable (*I broke the window by swinging a rock with my arm). A host of subevents leading to the breaking of a window are of course conceptually present. The preceding examples show that underspecification is an essential feature of language in use. In an ongoing piece of discourse linguistic expressions tend to evoke large amounts of knowledge. Fauconnier’s (1999: 96) wording on this issue deserves to be quoted in full: Language is only the tip of a spectacular cognitive iceberg, and when we engage in any language activity, be it mundane or artistically creative, we draw unconsciously on vast cognitive resources, call up innumerable models and frames, set up multiple connections, coordinate large arrays of information, and engage in creative mappings, transfers and elaborations. This is what language is about and what language is for.

In the spirit of this cognitive view of language, we propose the following definition of meaning construction: Meaning construction is an on-line mental activity whereby speech participants create meanings in every communicative act on the basis of underspecified linguistic units.


This definition does not restrict the notion of meaning construction to the creation of novel meanings, i.e. novel representations. The task of constructing meanings is probably more demanding when they are novel than when they are conventionalized, as will be shown in the examples below. A second point in the definition that deserves to be stressed is the role of the speech participants. The process of meaning construction is usually understood from the perspective of the listener while the role of the speaker is underestimated. However, both the speaker’s and the listener’s perspectives need to be taken into account because both are involved in meaning construction. In planning his linguistic expressions, the ideal speaker anticipates the particular meaning construction which the listener will presumably create. If there is a mismatch between the anticipated meaning construction and the one created by the hearer, the interlocutors will negotiate repair strategies in order to obtain a closer match between the two meaning constructions. Thus, meaning construction is a dynamic process in which fine-tuning between the interlocutors plays an essential role. The relevance of fine-tuning becomes particularly clear in situations of miscommunication. Let us use one of Deborah Tannen’s (1990: 51) examples between two speakers to illustrate the point.

(2) a. b. c. d.

he: I’m really tired. I didn’t sleep well last night. she: I didn’t sleep either. I never do. he: Why are you trying to belittle me? she: I’m not! I’m just trying to show that I understand.

The man apparently intended to communicate his nightly suffering (2a), and the woman constructed the illocutionary meaning from his underspecified utterance. Her reply (2b), however, gave rise to a serious misunderstanding: she intended to show her understanding whereas he interpreted her utterance to mean that she denied him “the uniqueness of his experience.” This example also demonstrates that meaning construction is an online process in which the communicative value of each utterance is determined and redefined with each turn and contribution. It also shows that meaning construction is not solely a matter of the hearer and his interpretation of an utterance; it crucially also involves the speaker and his assessment of the hearer’s state of mind, his knowledge, his emotions, etc. Only when we also consider the other person’s perspective can we . In this respect we take a different view of the notion of meaning construction from Evans and Green (2006: 363), who restrict their understanding of meaning construction to novel representations: “Meaning construction is the process whereby language ‘prompts for’ novel cognitive representations of varying degrees of complexity. These representations relate to conceived scenes and aspects of scenes, such as states of affairs in the world, emotion and affect, subjective experiences, and so on.”


detect the whole impact of meaning construction in the above dialogue. The man produced utterance (2a) with the expectation in mind that his spouse would understand the intended meaning of his utterance and pity him; the woman gave her reply (2b) in the belief that sharing the same experience in this particular situation would mean the same to her companion as it did to her, which was obviously not the case. Processes of meaning construction resulting from underspecification are, in principle, the same in miscommunication and successful communication. In his paper “Constructing the meaning of personal pronouns,” Langacker (this volume) draws attention to “intersubjectivity” in communication, i.e. the apprehension of other minds. Intersubjectivity involves the conceptual coordination between the interlocutors, who are simultaneously the conceptualizers.

2. Types of underspecification Linguistic underspecification is manifested in different ways. We see at least three ways in which linguistic units can be said to be underspecified and in need of conceptual augmentation: (i) linguistic units verbalize meanings implicitly or indirectly, (ii) linguistic units are indeterminate, and (iii) linguistic units are incompatible. These three manifestations of linguistic underspecification will be termed ‘implicitness,’ ‘indeterminacy,’ and ‘incompatibility.’ These types of underspecification do not establish clearly defined categories but may overlap.

2.1 Implicitness The notion of implicitness in this context is taken to mean that either no linguistic unit is present or that the linguistic unit is not conventionally associated with a particular meaning. This is the most common type of underspecification. Simple words and complex expressions underspecify conceptualizations in different ways. Simple words are associated with one or more conventionalized meanings which, within a given context, serve as prompts for the activation of richer conceptualizations (see also Nikiforidou, this volume). In the well-known case of the commercial event frame, any word denoting an aspect of a commercial event prompts the commercial event frame. Verbs such as buy, sell, cost, pay, etc. provide different entry points while the commercial event frame as a whole is not verbalized. At the same time, each entry point may further invite more specific processes of meaning constructions. For example, in Harry sold his car for a good price, the amount of money paid in this transaction is understood to be higher than in Harry bought his car for a good price. Again, this meaning is not present in the linguistic expression. A further example of implicit meanings evoked by a single word is the adjective safe in


usages such as The beach is safe. Fauconnier and Turner (2002: 25–7) show that safe prompts us to construct a counterfactual scenario of danger in which a person is harmed. None of this is present in the linguistic unit. Complex expressions tend to underspecify with respect to the individual lexical items and with respect to their composition. As argued by Langacker (2000: 152), the component structures of complex expressions usually motivate, but do not predict, the composite structure. Language thus exhibits partial rather than full compositionality, i.e. the meaning of a composite structure is underspecified and needs to be constructed. This applies to morphological, syntactic, pragmatic and discourse structures alike. Many scholars have shown that the meanings of compounds and derivations are underspecified from their components. Thus, the compound trouser suit does not indicate that it is women’s garment, the compound push chair does not reveal that it is designed to move children in it, and the compound wheelchair does not express that it is meant to transport elderly and/or sick people. Many syntactic constructions suppress entities that are part of the situation: Ruiz de Mendoza and Mairal Usón (this volume) have pointed out that, in nominalizations such as destruction, participants of the event may be left implicit; similarly, the inchoative construction in The door opened suppresses the controlling entity. Other examples of suppression include the passive construction, which optionally suppresses the agent, the imperative, which suppresses the addressee, modal verbs, which obligatorily suppress the speaker (*You may come in by me), complement clauses, which may suppress the subject participant, etc. Grammatical implicatures are also instances of meaning construction. As shown by Posner (1986), the ordering of attributive adjectives may invite implicatures. For example, a small precious stone is a precious stone that is small, while a precious small stone is a small stone that is precious. Furthermore, our world knowledge of precious stones tells us that a small precious stone is smaller than a precious small stone, none of which is explicated in language.

2.2 Indeterminacy Indeterminacy refers to situations in which a linguistic unit is underspecified due to its vagueness in meaning. Indeterminacy is, as suggested by Langacker (in press), the canonical situation in grammar. It is probably also the canonical situation with lexical categories. This point can be illustrated by an example taken from Talmy (1988: 92). The verb break is indeterminate with respect to the manner, the effect, the speed of breaking, etc. Thus, break in the context of sentences (3a) and (3b) evokes very different meanings of breaking due to our knowledge of the world:

(3) a. The heat broke the guitar. b. A falling radio broke the guitar.


The manner of breaking in sentence (3a) is seen as slow, leading to gradual warping and tracework of cracks, and the like, while the manner of breaking in sentence (3b) is seen as involving a sudden localized disruption. Grammatical indeterminacy is found in possessives (my horse can mean ‘the horse I own,’ ‘the horse I ride,’ ‘the horse I bet on,’ etc.), pronouns (we can include or exclude the hearer), attributive adjectives (nice in a nice present can be restrictive or non-restrictive), modal verbs (The car should have air conditioning is vague between an epistemic and root interpretation), indefinite quantifiers, and a host of other grammatical phenomena. In a study by Hörmann (1983, quoted in Gibbs 1994: 38f), subjects were asked to estimate the number of people expressed by a few in different situations. The lowest, the median and the highest average estimates are given in (4):

(4) a. a few people standing before a hut b. a few people c. a few people standing before the building

4.55 5.72 6.99

Gibbs (1994: 39) interprets these findings in terms of meaning construction: “Listeners do not simply look up the particular number associated with a few but construct its meaning, given some specific situation.”

2.3 Incompatibility Incompatibility is understood as an instance of underspecification since it requires the interlocutors to construct meanings in order to reconcile the conflict between expressions. These meanings are prompted by linguistic expressions but not linguistically specified. Studies on meaning construction have mainly been concerned with this type of underspecification, probably because its deviant linguistic structure and the listener’s effort after meaning call for an explanation. For obvious reasons incompatibilities are restricted to composite structures. Incompatibilities may occur between lexical units or between lexical and grammatical units—there do not seem to be incompatibilities between grammatical units. Lexical incompatibilities include the composition of antonyms, tautology, metaphor, metonymy, oxymoron, understatement, and profile/active-zone discrepancy; lexical-grammatical incompatibilities involve mismatches between a construction and lexical units used in it. A few of these instances of incompatibility will be dealt with here. People’s interpretation of antonymic adjectives was studied by Byrne (1979, reported in Posner 1986). Byrne asked his subjects to interpret noun phrases with contradictory attributive adjectives. The phrases were not rejected as meaningless but interpreted iconically according to the order of their adjectives. Thus, a slow fast dog was understood as a fast dog made temporarily slow, e.g. by injury, illness or age,


and a fast slow dog was interpreted as a slow dog made temporarily fast, e.g. a rocket-powered Saint Bernard. We interpret these findings in terms of meaning construction. The iconic principle of proximity provides an inferential pathway along which meaning construction may occur: units which belong together conceptually are more closely integrated syntactically. On the basis of this iconic principle, the more distant adjective invites aspects of meaning construction such as ‘be made,’ ‘temporarily’ or a setting in concrete real-world situations. Tautologies may also be seen as instances of lexical incompatibility: the predicate of a nominal tautology is not compatible with the subject because it predicates the very content of the nominal and hence is not informative in its literal meaning. This linguistic situation therefore prompts meaningful interpretations. Thus, the well-known conversational tautology Boys will be boys is, depending on the context, understood to mean ‘boys are unruly’ or ‘boys are cute and adorable’ (Gibbs 1994: 345–51). In this case the meaning construction prompted by the tautological expressions is guided by metonymic reasoning based on shared stereotypical knowledge. Metaphor, metonymy, oxymora, and some other “figures of speech” involve semantically “deviant” usages of language. Their incompatibility is resolved by constructing a target meaning which the speaker apparently intended to convey. Thus, in the sentence She is married to a library, the semantics of be married and library are incompatible and require the construction of meaning. The deviance may reside in either of the two terms. The listener may abandon the literal meaning of be married and construct a metaphorical meaning such as ‘frequent a library’ or ‘study at a library,’ or he may abandon the meaning of library and construct a metonymic meaning such as ‘extremely well-read person,’ ‘lover of books’ or ‘person who has books all over.’ In either case, the meaning construction is prompted by source expressions which are incompatible with other linguistic material which belongs to the target. Figures of speech other than metaphor and metonymy have hardly been studied within the cognitive-linguistic framework (apart from Gibbs 1994). The oxymoron in screaming silence and the understatement in She’s a bit of all right, meaning ‘good-looking’ or ‘terrific,’ of course also require substantial efforts to construct their meanings. Probably the commonest—and hence least noticeable—type of incompatibility is the phenomenon described by Langacker as “profile/active-zone discrepancy.” An entity in a relationship is expressed, or profiled in Langacker’s terminology, but only a facet of it, an active zone, directly and crucially participates in this relationship. In his example The cigarette in her mouth was unlit, only “a particular portion of the cigarette (one end) was contained in a particular portion of the mouth (a segment of the lips)” (Langacker, in press). Profile/active zone discrepancy is a kind of metonymy, i.e. whole for part, and involves focusing on an active zone as


its meaning construction, something which is normally performed automatically within a culturally given context. Incompatibilities between grammatical constructions and lexical units are often treated within the framework of construction grammar. The meaning of the grammatical construction may sometimes be in conflict with the meaning of lexical items which are not normally used in this construction. Probably the best-known study of a lexico-grammatical incompatibility is Goldberg’s (1995) analysis of the caused-motion construction with verbs that do not denote motion. In her classic example Fred sneezed the napkin off the table, the caused-motion construction in conjunction with the lexical verb sneeze prompts a construction of meaning which can be paraphrased as ‘with the force of one’s sneezing cause something to move away in a sudden, explosive, etc. manner.’ What is underspecified in this sentence is not the construction but the lexical item. The caused-motion construction imposes a shift in the grammatical valency of the verb from a one-place predicate to a three-place predicate and concomitantly a shift in meaning from a non-causative meaning of the verb to a caused-motion meaning. The verb sneeze in this usage is thus underspecified with respect to the causal nature of the subject argument, an argument describing an affected entity and a further argument describing a direction. In addition, other aspects of meaning are inferred such as ‘abrupt and uncontrolled motion’ and ‘motion away from the sneezer.’ The following section will deal with possible ways of resolving this and other types of underspecification.

3. Ways of constructing underspecified meanings The three types of underspecification distinguished in the preceding section share the property that underspecified linguistic units evoke a whole scenario. We think and experience the world in terms of scenarios. Linguistic units are prompts for accessing such scenarios, i.e. the meanings of these linguistic units metonymically stand as parts for the whole scenario. The part for whole-metonymy can thus be seen as a fundamental principle of all kinds of meaning construction. Scenarios are naturally part of a given culture and are crucial sources of information in meaning construction. Without knowledge of the scenarios surrounding September 11, 2001, language users would not be able to understand the meaning of a sign in public buses in New York City: “If you see something, do something.” The linguistic units in this sentence are at the highest possible level of generality, or indeterminacy, and could mean almost anything: the indefinite pronoun something neither specifies what you have to look for nor what one has to do. Yet most Americans will probably interpret this highly underspecified message in precisely the way it was meant to be understood.

10


Meaning construction is an inferential process. The potential range of inferences is constrained by the linguistic expressions that serve as prompts, the situational and discourse context, the language users’ encyclopaedic knowledge, their personal attitudes and certainly many more factors. These are very specific determinants and account for specific aspects of meanings. It is at this specific level that potentially different scenarios are evoked, as in the case of Boys will be boys, which may give rise to scenarios of ‘boys’ unruly behaviour,’ ‘their being cute’ or ‘their being adorable.’ Whichever scenario is evoked as the listener’s meaning construction, it is based on a general principle according to which things may stand for one or some of their properties. The conceptual metonymy thing for property of a thing is readily available to members of a community and guides more specific inferences. For example, the presenter who completed her first public talk at the ICLC in Seoul by saying “I’m just a student” did, of course, not want to inform the audience about her academic status. Thanks to our knowledge of the thing for property of the thing-metonymy we know that her underspecified concluding words were meant to refer to properties of students. For lack of any evidence to the contrary we think of properties set in stereotypical scenarios associated with students in the Western culture: ‘intelligent,’ ‘partially trained academically,’ ‘critical,’ etc. Since her utterance was situated in an academic setting and meant as a meaningful act of communication, we further infer that its illocutionary force was probably that of an apology in case the audience felt that, due to her partial academic training, her paper might not meet with the high academic standards expected of presentations. In what follows we will mainly concentrate on the general principles guiding meaning construction like the conceptual metonymy in the above examples. Such general guiding principles of meaning construction are metonymy, metaphor, conceptual blending, iconicity and possibly some more. The papers in this volume concentrate on the first three principles in various aspects of meaning construction. Panther and Thornburg (2003, 2004) were among the first to recognize the general function of conceptual metonymy in inferences. They describe metonymy as natural inference schemas and inferential pathways. Conceptual metonymies operate at an intermediate level and guide the interlocutor in the construction of meaning. Their metonymic approach to meaning construction has been especially insightful in their analysis of lexico-grammatical incompatibility, i.e. in cases of grammatical coercion. For example, a number of constructions are only compatible with actions but are often used with stative predicates, like the imperative construction in Be wealthy in ten months. The meaning of the sentence can be paraphrased ‘Do something to the effect that you will be wealthy in ten months.’ The schematic meaning to be constructed is that of an action leading to a resultant state. Actions and their results are related metonymically, and our understanding of the sentence makes use of the widely used metonymy result for action. This me-


tonymy serves as a general inferential pathway that guides the listener to more specific interpretations. The specific kinds of action to be taken have to be inferred by reasonable guesses, such as the purchase of shares. Traditionally, it is claimed that grammatical constructions override the meanings of lexical expressions. Panther and Thornburg (2004), however, show that coercions are not always unidirectional. In their example Enjoy your summer vacation! the imperative construction does not, in its non-directive reading, coerce lexical adjustment, but the lexical item enjoy with its meaning of ‘spontaneously occurring experiential state’ coerces an adjustment of the meaning of the imperative construction: it is now understood to express an optative meaning, namely the speaker’s wish or hope. Metonymy is without doubt a powerful principle guiding inferential processes of meaning construction. Many of the contributions in this volume have been inspired by Panther and Thornburg’s view of metonymies as natural inference schemas. For example, Ruiz de Mendoza Ibáñez and Mairal Usón focus on high-level metonymies in meaning construction. High-level metonymies are particularly pertinent in grammar. For example, they interrelate actions, processes and results. Thus, the process use of close in The door closed well thanks to her instructions involves the metonymy process for action, and the middle construction in This bread cuts easily is motivated by the double metonymy process for action for result. Like conceptual metonymies, conceptual metaphors serve as pathways for meaning construction. As shown by Lakoff and Johnson (1980: 52–5), the conceptual metaphor theories are buildings is structured by focusing on certain parts of a building (foundations and outer shell) but also allows us to exploit other parts (the roof, rooms, staircases, hallways, etc.) which are normally not used in the metaphor. When we are confronted with a sentence like His theory is covered with gargoyles, we understand its schematic meaning on the basis of the metaphor theories are buildings. We can then make further reasonable guesses about this particular theory by relating it to a scenario like that of grotesque creatures sitting on the gutters of medieval churches. Image metaphors, i.e. metaphorical expressions which are not based on conceptual metaphors, are not expandable in the same way. For example, the word bicycle is used in soccer to describe a player’s shot while falling backwards. This metaphorical term does not motivate the use of unused parts of a bicycle like the saddle, the handle-bars or the chain in talking about soccer. The lack of a conceptual metaphor as an inferential pathway may also increase the difficulty of constructing an adequate meaning. Thus, a trite sentence such as Philosophy is the foundation of our knowledge is easily interpretable due to our knowledge of the theories are building-metaphor, but the graffito Philosophy is the wallpaper of the mind requires quite some “effort after meaning” and will probably be interpreted differently by different people.

11

12


Conceptual metaphor is a predominantly lexical inference schema and less commonly found in grammar. Ruiz de Mendoza Ibáñez and Mairal Usón (this volume) provide instances of high-level metaphors operating at the lexico-grammatical level. One of their examples is the metaphor communicative action is effectual action, which accounts for the conversion of the intransitive verb talk (to someone) to a goal-directed transitive verb in He talked me into it. Variations in the use of the caused-motion construction with intransitive verbs are extensively discussed by Wulff, Stefanowitsch, and Gries (this volume). The notion of meaning construction has mainly been associated with mental spaces and the emergence of new meanings by conceptual blending. Fauconnier and Turner (1996: 40) define mental spaces as “small conceptual packets constructed as we think and talk, for purposes of local understanding and action.” Conceptual blending occurs when two or more mental spaces are conceptually integrated to form a new complex concept. This is, for instance, the case in morphological blends and compounds, whose complex meanings are derived from blending two input spaces. In Fauconnier and Turner’s (2000) example landyacht, selective elements of the two named input spaces ‘land’ and ‘yacht’ are projected onto a blended space and give rise to the emergent meaning ‘expensive luxury car.’ Conceptual blending also occurs from implicit cues as in the case of the danger frame evoked by safe in 2.1. A highly underspecified space-builder is also the negator no in a sentence like There is no beer in the fridge. The meaning construction involves “unpacking” the blend composed of a reality space in which there is no beer in the fridge and a counterfactual space triggered by no in which there is beer in the fridge. The input spaces are typically related by a small set of vital relations, which serve as general principles guiding meaning construction, such as relations of time, space, representation, cause-effect, analogy, and reality-counterfactuality. Relations between mental spaces may also involve ‘contiguity’ or ‘similarity’ and hence are metonymic or metaphoric, i.e. metonymy and metaphor may be seen as special instances of conceptual blending (see Coulson and Oakley 2003 for metonymic blends and Grady 2005 for metaphorical blends). The contributions collected in this volume are grouped according to their focus on meaning construction in metonymy and metaphor on the one hand and mental spaces and conceptual blending on the other hand. Part I is devoted to meaning construction in metonymy and metaphor. The papers address the following issues: Raymond Gibbs, Jr. surveys recent research on online meaning construction related to figurative language. While empirical research from psycholinguistics has not yet been able to investigate the conceptual nature of metonymy, there is considerable evidence of metonymic inferencing in language understanding and meaning construction. Francisco José Ruiz de Mendoza Ibáñez and Ricardo Mairal Usón demonstrate


the role of high-level metaphor and metonymy in the construction of meaning in the domain of transitivity. Antonio Barcelona studies meaning construction through implicatures in a piece of narrative discourse. The results of his empirical investigation show that readers’ expectations about the development of the text are to a large extent guided by metonymies. Martin Hilpert investigates the construction of meaning through chained metonymies. His cross-linguistic study is concerned with body part terms and their lexical extensions and development of grammatical meanings. Debra Ziegeler argues that the notion of coercion is an inadequate tool in accounting for mismatches at the syntax–semantics interface. In her discussion of three representative cases of “coercion” she proposes alternative explanations, in particular metonymy, which are better suited to account for the construction of meaning. Mario Brdar and Rita Brdar-Szabó study the use of figuratively used personal names and show how their online construction of figurative meaning involves tiers of metonymic and metaphoric mappings. Anatol Stefanowitsch investigates the collocations in which selected metaphors occur and, on the basis of rich corpus data, concludes that statistical properties of collocational overlap can guide the construction of metaphoric meaning. Part II is devoted to mental spaces and conceptual blending in the construction of meaning. It includes the following contributions: Ronald Langacker explores the conceptual function of the personal pronouns I and you. The abstracted meanings of these pronouns invoke very basic cognitive models pertaining to the ground, a speech event, subject vs. object of conception, and intersubjectivity, which refers to the interlocutors’ mutual apprehension of each other’s minds. Kiki Nikiforidou shows that Greek relatives may be highly underspecified with respect to the intended interpretation and, in order to be interpreted, require the construction of a blend with complex emergent meanings. Christian Koops addresses the cross-linguistic variability of inferential constructions, in particular the it is that-construction, a construction that facilitates the hearer’s construction of meaning. He shows that this construction is highly restricted and constrained in discourse. Wiltrud Mihatsch studies the emergence of Romance approximators such as espèce de, which are derived from taxonomic nouns like espèce ‘species.’ This shift in meaning is motivated by communicative goals related to loose meaning construction. Wolfgang Schulze argues that questions are the immediate reflexes of a memory mismatch. Questions require a change of perspective and are seen as alter-ego

13

14


conceptualizations, a notion that is comparable to Langacker’s notion of intersubjectivity. Stefanie Wulff, Anatol Stefanowitsch, and Stefan Gries show that the into-causative construction has a much wider application in British English than in American English and hence is associated with different meaning constructions in these two varieties of English.

References Bartlett, Sir Frederic C. 1932. Remembering: A Study in Experimental and Social Psychology. Cambridge: Cambridge University Press. Reissued 1995. Byrne, Brian. 1979. Rules of pronominal adjective order, and the interpretation of incompatible adjective pairs. Journal of Verbal Learning and Verbal Behavior 18: 73–8. Coulson, Seana and Oakley, Todd. 2003. Metonymy and conceptual blending. In Metonymy and Pragmatic Inferencing, Klaus-Uwe Panther and Linda L. Thornburg (eds), 51–79. Amsterdam, Philadelphia: Benjamins. Evans, Vyvyan and Green, Melanie. 2006. Cognitive Linguistics: An Introduction. Edinburgh: Edinburgh University Press. Fauconnier, Gilles. 1999. Methods and generalizations. In Scope and Foundations of Cognitive Linguistics, Theo Janssen and Gisela Redeker (eds), 95–127. The Hague: Mouton de Gruyter. Fauconnier, Gilles and Turner, Mark. 2002. The Way We Think: Conceptual Blending and the Mind’s Hidden Complexities. New York: Basic Books. Gibbs, Raymond W., Jr. 1994. The Poetics of Mind: Figurative Thought, Language, and Understanding. Cambridge: Cambridge University Press. Goldberg, Adele. 1995. Construction: Construction Grammar Approach to Argument Structure. Chicago: Chicago University Press. Grady, Joseph. 2005. Primary metaphors as inputs to conceptual integration. Journal of Pragmatics 37: 1595–1614. Hörmann, Hans. 1983. The calculating listener or how many are einige, mehrere, and ein paar (some, several, and a few). In Meaning, Use, and Interpretation of Language, Rainer Bäuerle, Christoph Schwarze, and Arnim von Stechow (eds), 221–34. Berlin: de Gruyter. Lakoff, George and Johnson, Mark. 1980. Metaphors We Live by. Chicago: Chicago University Press. Langacker, Ronald W. 1987. Foundations of Cognitive Grammar, Vol. 1: Theoretical Prerequisites. Stanford, CA: Stanford University Press. Langacker, Ronald W. 2000. Grammar and Conceptualization. Berlin, New York: Mouton de Gruyter. Langacker, Ronald W. In press. Metonymy in grammar. In Metonymy and Metaphor in Grammar, Linda Thornburg, Antonio Barcelona and Klaus-Uwe Panther (eds). Panther, Klaus-Uwe and Thornburg, Linda. 2003. Metonymies as natural inference and activation schemas. In Metonymy and Pragmatic Inferencing, Klaus-Uwe Panther and Linda Thornburg (eds), 127–47. Amsterdam: Benjamins.


Panther, Klaus-Uwe and Thornburg, Linda L. 2004. The role of metonymy in meaning construction. metaphorik.de 6: 91–116. Posner, Roland. 1986. Iconicity in syntax: The natural order of attributes. In Iconicity: Festschrift for Thomas A. Sebeok, Paul Bouissac, Michael Herzfeld, and Roland Posner (eds), 305–37. Tübingen: Stauffenburg. Talmy, Leonard. 1988. Force dynamics in language and cognition. Cognitive Science 12: 49–100. Talmy, Leonard. 1996. The windowing of attention in language. In Grammatical Constructions: Their Form and Meaning, Masayoshi Shibatani and Sandra A. Thompson (eds), 235–87. Oxford: Clarendon Press. Tannen, Deborah. 1990. You Just Don’t Understand: Women and Men in Conversation. New York: Ballantine Books.

15

part 1

Metonymy and metaphor

chapter 1

Experimental tests of figurative meaning construction Raymond Gibbs, Jr. University of California, Santa Cruz

One of the major claims of recent cognitive linguistics research is that metonymy constitutes a fundamental scheme of human cognition and is not just a rhetorical device employed for specific communicative purposes. The work of Klaus-Uwe Panther and others has suggested that certain metonymies are natural inference schemes operating during many aspects of language production and understanding. This chapter explores the relations between cognitive linguistic ideas on conceptual metonymy and recent psycholinguistic experiments examining online meaning construction. I suggest that there is no direct evidence supporting the idea that conceptual metonymies are immediately recruited during metonymic language processing, but that this gap is due to the difficulties in testing whether very abstract schemes are accessed during online meaning construction. Nonetheless, there exists various experimental support for other cognitive linguistic claims about metonymy, including the importance of metonymy for highlighting certain aspects of discourse topics, the interaction of metonymy and grammatical structure in sentence comprehension, and the idea that conceptual metonymies may interact with pragmatic information to constrain specific interpretations of metonymic utterances. Keywords: conceptual metonymy, corpus linguistics, figurative language, meaning construction, metaphor, metonymy, pragmatics, psycholinguistic experiments

1. Introduction Some time ago I was in a local pub with friends when the following brief conversation occurred as a late arrival approached the group. Greg: “Hi, is Nicole still here?” Megan: “She left about two beers ago.” When Megan gave her reply to Greg’s question, everyone present seemed to have a reasonable understanding about the approximate time that Nicole left the pub (about 30 minutes ago) as I immediately inquired as to how they interpreted Megan’s indirect response. How did this general understanding of Megan’s utterance happen? The phrase “two beers ago” is not typically used to indicate a period of

20

Raymond Gibbs, Jr.

time, yet in this context people were readily able to infer that “two beers ago” referred to a period of time in which the last two beers were consumed. Listeners used their pragmatic knowledge of the local situation (e.g., the amount of time in which two beers had been consumed), in addition to conceptual information that “two beers ago” could stand for the time it took everyone to drink the last two beers. This conceptual knowledge was not just specific to the drinking of beers, but more generally suggests how the mention of some activity can stand for the time it takes for that activity to occur (i.e., activity for time to do that activity). In this way, our understanding of “She left about two beers ago” relied on our ability to quickly access both specific pragmatic knowledge and a conceptual metonymy and integrate these two kinds of information to construct an appropriate figurative reading of Megan’s comment. My aim in this chapter is to discuss some of the relevant empirical work on how pragmatic and conceptual information are accessed and mutually constrain people’s understanding of figurative meaning, with a special emphasis on metonymy. I specifically explore the relation of ideas from cognitive linguistics on metonymy to experimental work in psycholinguistics. The key question here is: Do theories from cognitive linguistics provide psychologically real accounts of how people construct meaning in ordinary discourse situations? My description of empirical works suggests a positive answer to this question. Nonetheless, understanding metonymy requires important mutual adjustment processes accessing both conceptual and pragmatic information during online meaning construction.

2. Metonymy and meaning construction The idea that conceptual metonymy plays a central role in ordinary meaning construction has been proposed in recent work by Panther and Thornburg (2003, 2004). They distinguish their account of conceptual metonymy from pragmatic theories of utterance interpretation, such as Grice’s (1989) and that of relevance theory (Sperber and Wilson 1995), which assume that more abstract inferential processes, based on maxims or the principles of relevance, are required for listeners to derive appropriate meaning for what speakers say. A significant part of Panther and Thornburg’s claim is that metonymy provides what they refer to as “natural inference schemes,” because these schemas reflect recurrent, entrenched conceptual mappings. Conceptual metonymies, such as part for whole, cause for effect, person for role, and representation for represented, are not just used in inferential communication, but are “vital relations” underlying reasoning. Conceptual metonymies reflect an intermediate level of inferential reasoning, precisely because they are sufficiently abstract to serve a wide variety of inferential situations, and yet are

Figurative meaning construction

specific enough to provide detailed accounts of meaning construction in specific contexts of language use. They determine both explicit meaning (i.e., explicatures) and implicit communication (i.e., particularized conversational implicatures). Recent work on conceptual metonymy has done much to correct the overbearing privilege of metaphor in cognitive linguistic studies of language and thought. Especially the collections of papers in Panther and Radden (1999) and Panther and Thornburg (2003) have presented detailed illustration of the ways that metonymy shapes language, thought, and more specifically, pragmatic inferencing. The advocacy of conceptual metonymy over strict pragmatic theories of utterance interpretation by many cognitive linguists does not imply that pragmatics has no role in meaning construction. On the contrary, as correctly claimed by Panther and Thornburg, conceptual metonymy provides an important constraint on pragmatic inferential processes, and that conceptual factors are clearly integrated with pragmatic ones in utterance interpretation. For example, they have shown that conceptual metonymies constrain pragmatic functions such as the referential, predicational, and illocutionary levels of speech acts, as well as shape the organizational content of the mental lexicon, interact with grammatical structure, and are employed in the creation of novel pragmatic meaning. Presented below are my reflections on the psychological status of several cognitive linguistic claims about conceptual metonymy and I will suggest some concrete ways that conceptual metonymies interact with pragmatic information to create specific, contextually-appropriate inferences from speakers’ utterances in conversation.

3. Proving the psychological reality of conceptual metonymies Are conceptual metonymies such as part for whole, object for user, and place for event really conceptual in the sense that people have enduring knowledge of such relations that affects their immediate thoughts and language use? Linguistic evidence is, at the very least, suggestive of the enduring cognitive, and not just linguistic, nature of metonymy. This does not imply a secondary status for the linguistic examination of metonymy in comparison to investigating metonymy as a pervasive scheme of thought. But cognitive linguistics argue that metonymy is fundamentally motivated by thought in the sense that people often think metonymically, which provides the grounding for why speakers use metonymy. “Metonymy is speech” is therefore not just a rhetorical move that people engage in for special communicative purposes, as traditionally thought within linguistics and literature (Gibbs 1994). But the cognitive linguistic work gives rise to the question as to whether there is specific evidence that people actually have generalized conceptual mappings such as object for user, and recruit this knowledge as they think, speak, and under-

21

22

Raymond Gibbs, Jr.

stand. Do people speak metonymically as frequently as suggested by cognitive linguistic analyses (see the essays in Panther and Thornberg 2003)? Most cognitive linguistic studies have examined a limited range of texts in their exploration of metonymic reasoning. Panther and Thornburg, to their credit, have at least investigated a diverse set of examples in making their claims for the prominence of conceptual metonymy in language and thought. It is frankly unclear from both a linguistic and psychological perspective as to how frequent a particular type of figurative expression must be within a language for it to be recognized as a significant part of the way people ordinarily speak and think. Nonetheless, there is a growing movement in figurative language scholarship to conduct corpus studies that aim to provide actual counts of the distribution of metonymic language. One study examined the British National Corpus using an annotation scheme that detected metonymic regularities in regard to country names and reliably distinguished literal from metonymic usage (Markert and Nissim 2003). Contrary to the claim that metonymies referring to place for event (e.g., “He was shocked by Vietnam”) are frequent in discourse, less than 1% of all country names were used to refer to events. Sixteen percent of the country names metonymically referred to people (i.e., place for people), and there were no instances found related to place for product. But a further analysis showed that countries which were economically and politically more important to British readers, such as Japan, had a significantly higher proportion of metonymic readings than did less important countries (e.g., Guyana). This higher proportion of metonymic interpretations of more relevant, important country names was primarily due to the higher number of place for person metonymies. Corpus linguistic research like this is important for more clearly establishing the frequency and specificity of figurative meaning constructions. At the very least, such work provides an important empirical complement to linguistic research based on individual scholars’ own intuitions and created examples. Yet empirical counts of certain kinds of language use, like that of Markert and Nissum (2003) suggest that cognitive linguists must be very cautious before claiming that any particular figure of speech is common. The fact that people can possibly and meaningfully talk about place for event metonymies does not necessarily imply that speakers indeed frequently produce such figures in spoken or written language. Further discussion within cognitive and corpus linguistics needs to address the criteria for making theoretically motivated claims about the ubiquity of metonymy, and metaphor for that matter, in real discourse. For example, how frequent does a particular metonymic mapping have to be in language to indicate the presence of an underlying conceptual metonymy motivating this speech? Corpus linguistic studies of metonymy can also be employed to provide critical tests of specific ideas about conceptual metonymy, such as Panther and Thornburg’s


claim that in a prototypical metonymy, the target meaning is more prominent than the source. Thus, expressions such as “He was shocked by Vietnam” are readily understandable as having a metonymic reading because of the prominence of the specific target (place). In fact, as Panther and Thornburg (2004) argue, the traditional view of metonymy as a substitution for some other term may have some validity in cases where the target meaning has become “maximally prominent,” exactly as is the case in an expression like “He was shocked by Vietnam.” But the substitution works precisely because of people’s highly specific conceptual understanding of Vietnam as a concrete instantiation of the more general place for event. Proving the conceptual reality of metonymy, as with metaphor, is very much dependent on the results of psychological studies. There is a sizeable literature in experimental psycholinguistics demonstrating that conceptual metaphors (e.g., life is a journey) are at least used when individuals make sense of why many conventional and novel metaphors have the particular meanings they do (Gibbs 1994, 2006). Conceptual metaphors may also be recruited during people’s immediate comprehension of much metaphoric language (Gibbs et al. 1997; Pfaff, Gibbs, and Johnson 1997), although there remains significant debate over this issue (see Glucksberg 2001). In contrast to the plethora of studies on conceptual metaphor, there are virtually no experimental studies that have explicitly attempted to find evidence on the role of conceptual metonymies in figurative meaning construction. I have previously argued that various evidence from both cognitive psychology and psycholinguistics is quite consistent with the idea that people normally engage in metonymic reasoning during many aspects of ordinary text and spoken language understanding, especially that referring to part-whole relationships (Gibbs 1994, 1999). But, once again, there are no studies that show that those specific conceptual mappings, referring to basic metonymic relations, shape online meaning construction. Is there a reason for psycholinguists’ failure to study conceptual metonymies in figurative meaning construction, despite the prominence of work by cognitive linguists such as Panther demonstrating the salience of metonymy in many aspects of utterance interpretation? One difficulty in experimentally testing for the presence of conceptual metonymies during utterance interpretation is that these presumably entrenched knowledge structures are rather abstract. Consider first the situation with conceptual metaphor. Experimental studies have shown, for instance, that people appear to have tacit knowledge of conceptual metaphors (e.g., anger is heated fluid in a container) and use these to makes sense of many figurative expressions (e.g., “John blew his stack”). For example, people are faster to process metaphoric statements if these are read in contexts that depict similar metaphorical mappings (Pfaff, Gibbs, and Johnson 1997), and appear to draw inferences about what specific metaphoric statements mean based on the correspondences arising

23

24

Raymond Gibbs, Jr.

from underlying conceptual metaphors (Gibbs 1992). Furthermore, several studies show that people’s mental images for idiomatic and proverbial phrases in English are constrained by conceptual metaphors that link these phrases with their figurative meanings (Gibbs and O’Brien 1990; Gibbs, Strom, and Spivey-Knowlton 1997). Finally, exploring people’s intuitions about specific source domain information, such as that of hunger in the primary metaphor desire is hunger, enables researchers to make predictions about the kinds of source-to-target domains mappings that motivate the existence and understanding of many metaphoric expressions. For instance, the fact that people have certain embodied experiences of hunger that are particularly salient suggests that these should be mapped onto any talk about abstract desires unrelated to hunger. Indeed, people from two cultures, the United States and Brazil, rate statements strongly associated with salient hunger experiences (e.g., “He hungered for fame and fortune”) to be more acceptable than were phrases that reflected less salient aspects of embodied hunger (e.g., “I become talkative for fame and fortune”) (Gibbs, Lima, and Francuzo 2004). This research is important because it independently examines people’s embodied experiences and uses this information to make predictions about the kinds of figurative expressions ordinarily used to express metaphorical conceptions, such as those related to desire is hunger. Cognitive linguists work more deductively by examining systematic patterns of language to infer possible conceptual structures motivating such language. Yet they are never able to independently assess conceptual structures and the possible embodied foundations for those structures, because they are primarily working with linguistic materials. It is important to note that any experimental claim that conceptual metaphors (e.g. love relationships are journeys) are recruited in understanding conventional or novel metaphoric language (e.g., “Our marriage has hit a dead-end street”) does not imply that these conceptual metaphors provide all the knowledge needed to understand the figurative expression. Conceptual metaphors may only provide partial scaffolding for the eventual meaning constructed for what a person communicates with various sorts of enriched pragmatics having an important complementary role (see section 5). But this discussion of the experimental work on conceptual metaphor in figurative language understanding suggests that the specificity of conceptual and primary metaphors makes these entrenched knowledge structures more amenable to empirical investigation than appears to be the case for more abstract conceptual metonymies such as part for whole, object for user, place for event, or cause for effect. For example, it has been quite possible to ask people in various ways about their experiences of the source domain heated fluid in a container (i.e., anger is heated fluid in a container) or even more general experiences such as journey (i.e., life is a journey) in a way that seems quite difficult to do with


abstract domains like object, place, event, and so on. Psycholinguistic studies have shown, for instance, that people can immediately judge that the letter string “heat” is a word right after reading “John blew his stack,” because of the quick recruitment of the conceptual metaphor anger is heated fluid in a container (Gibbs, Bogdonovich, Sykes and Barr 1997). But it is not clear that the same type of study could be done with a target word like “user” when people read a metonymic phrase such as “The sax has the flu today” (i.e., object for user). The challenge for both cognitive linguists and psycholinguists is to discover more specific metonymic relations that may be amenable to empirical test and to find ways of showing that schematic knowledge of the sort like object for user seems to be used in speaking and listening. One possibility for the latter is to employ a priming paradigm to see if reading metonymic phrases motivated by specific conceptual metonymies, such as “The sax has the flu today” (object for user), facilitates people’s reading of metonymic phrases supported by similar conceptual metonymies (e.g., “We need a new glove at third base”) as opposed to phrases motivated by different conceptual metonymies (e.g., “Iraq is becoming another Vietnam” or place for event).

4. Studies on metonymic processing Despite the paucity of experimental studies that directly explore the possible role of conceptual metonymy in language use, psycholinguists have investigated a few features of metonymic language understanding, many of which are consistent with Panther and Thornburg’s own analyses. Experimental results show that people can readily understand referential metonymic expressions, such as “The scalpel was sued for malpractice” (in reference to a surgeon) as well as they do comparable referential metaphoric statements, such as “The butcher was sued for malpractice,” even if both types of statements take longer to process than non-figurative equivalents such as “The surgeon was sued for malpractice” (Gibbs 1990). One may argue that people’s relatively quick understanding of referential metonymies, at least in the sense of taking no longer to process than comparable metaphoric statement, is due to the immediate activation of a relevant conceptual metonymy, such as object for user in “The scalpel was sued for malpractice.” But these results per se do not directly show this to be the case, nor were they designed to do so. Also note that the result showing that both metonymic and metaphoric referential statements took longer to interpret than did corresponding literal expressions is likely due to the fact that the critical word indicating a metonymic or metaphoric meaning was in a sentence-initial position. I raise this point because cognitive linguistic studies of figurative meaning construction, most generally, often talk about

25

26

Raymond Gibbs, Jr.

how different utterances are interpreted by looking at the expression as a whole, but do not seek to explain the online, incremental nature of figurative meaning construction as people hear or read each word. Thus, part of the challenge in any theory of meaning construction is to describe exactly what sorts of conceptual and pragmatic information is employed and when it is accessed during immediate understanding. More specifically, reading the NP “The scalpel” may not invite any metonymic reading until the VP “was sued for malpractice” is encountered. In other cases, a metonymic reading may be constructed immediately upon seeing the target term as in “Susan enjoyed the book” (i.e., reading the book). Exactly when metonymic readings are accessed or constructed is clearly a topic for further psycholinguistic investigations that can examine online processing in an incremental manner. Finally, other aspects of the Gibbs (1990) findings provide empirical support for one important claim of Panther and Thornburg (2004) when they assert that metonymies both make targets accessible (e.g., “scalpel” activates the referent of “surgeon”) and available for further elaboration in ensuring discourse. When people were asked to describe why the initial noun phrase in each final statement (e.g., “The scalpel was sued for malpractice”) referred to its earlier stated referent (i.e., why the surgeon was called a “scalpel”), participants 88% of the time described the underlying nature of the metonymic relationship (e.g., “the surgeon was called a ‘scalpel’ because surgeons often used scalpels” or “the athlete was called a ‘glove’ because gloves are an important part of a baseball player”). This finding suggests that readers determine the figurative connections between the final metonymic statements and their previously stated referents, and do not simply associate a target with its referent in an arbitrary manner. Moreover, when participants were asked to talk more about the “surgeon,” for example, their discussions were primarily focused around the specific metonymic label (e.g., “the surgeon was just an uncaring object”). In this way, metonymic referential descriptions appear to make targets both accessible and available in discourse processing, just as argued by Panther and Thornburg (2004). A different set of experiments showed how people’s familiarity with specific targets facilitated metonymic processing. An analysis of people’s eye-movements as they read showed that processing of familiar place for institution metonymies such as “the convent” in “That blasphemous woman had to answer to the convent,” took no longer than did sentences containing literal interpretations of the same expressions (Frisson and Pickering 1999). On the other hand, metonymic statements that contained unfamiliar place for institution metonymies, such as “That blasphemous woman had to answer to the stadium,” took considerably longer to read than both familiar metonymic statements and literal equivalents. Similar findings were also reported for familiar and unfamiliar place for event metonymies, such as “A lot of Americans protested during Vietnam” (familiar) and “A lot of Ameri-


cans protested during Finland” (unfamiliar), although the processing delay for the unfamiliar statements arose late in their processing. Frisson and Pickering (1999) interpret these various results as being inconsistent with literal-first or figurativefirst models of figurative language processing. They alternatively argue that their findings support an “underspecification model” in which a single, underspecified representation for these metonymies arose first that was later fleshed out into a contextually appropriate figurative meaning. One may also conclude from Frisson and Pickering’s work that a reader’s familiarity with the specific instantiation of a conceptual metonymy has a clear influence on the speed with which people understand metonymic statements. For instance, when people read “A lot of Americans protested during Vietnam,” their correct understanding of this phrase is not completed once an abstract place for event metonymy is accessed or inferred, as readers must recognize what aspect of the place has a specific relationship to some event (e.g., what aspect of the country Vietnam was associated with American protests). To the extent that people have more familiar or stereotypical associations with a particular place, in this example, then the easier it will be for them to correctly construct what is meant by a metonymic statement like “A lot of Americans protested during Vietnam.” One possibility is that when a person hears an unfamiliar place for event metonymic statement such as “A lot of Americans protested during Finland,” they may immediately realize that some sort of place for event association is being drawn, and have this schematic representation in mind, but then experience difficulty determining the exact meaning of the expression due to a failure in seeing the specific instantiation of this conceptual metonymy (e.g., what event is saliently related to Finland?). Frisson and Pickering do not discuss the idea that an abstract conceptual metonymy is immediately recruited when reading a metonymic statement that is then fleshed out to provide more precise figurative meaning. Of course, eye-movement data alone are unable to detect the presence of underlying conceptual representations during utterance interpretation. Once again, what is needed are other experimental methods that enable researchers to see if specific kinds of conceptual representations arise during online meaning construction. Panther and Thornburg have argued that conceptual metonymy interacts with grammatical structure during utterance interpretation. Indeed, there is experimental evidence that is consistent with this idea. One eye-tracking study examined the processing of metonymies in syntactically ambiguous sentences (Pickering and Traxler 1998). For instance, participants read a context sentence mentioning either an actual professor or a statue of a professor, and then read a syntactically ambiguous sentence about a janitor polishing the professor on one syntactic analysis. The results showed that readers quickly adopted the plausible metonymic interpretation when it was available, which then affected the process of syntactic ambiguity

27

28

Raymond Gibbs, Jr.

resolution. Participants appear to have rapidly adopted the metonymic reading though no effects occurred at the point at which “professor” was first encountered. The point at which a metonymic reading is established during online meaning construction may differ depending on various lexical, grammatical, and pragmatic factors. Once more, showing exactly when metonymic readings are constructed, and whether these are influenced by entrenched conceptual metonymies is an important topic for figurative language research.

5. Mutual adjustment during figurative language processing Panther and Thornburg’s advocacy of conceptual metonymies still acknowledges the importance of pragmatics in figurative meaning construction. Let me suggest some specific ways this may work. Consider first the metaphor “Robert is a bulldozer.” How do we get from properties of machines to properties of human beings, a transition that is required in this instance of ad hoc concept construction? Does the context in which this statement appears shape its processing? In “Robert is a bulldozer,” the word “bulldozer” typically points to a set of familiar conceptual metaphors, such as people are machines, the mind is a machine, thinking is a mechanical activity, and so on. In some instances, the link between a lexical concept and a specific conceptual metaphor may be so entrenched that a specific metaphorical concept is immediately accessed, and found to help establish a relevant reading of “Robert is a bulldozer.” In other cases, listeners will need to test various conceptual metaphors in order of their accessibility until finding one that creates an interpretation of “Robert is a bulldozer,” which, as Sperber and Wilson (1995, also see Carston 2002) have shown, satisfies the expectation of optimal relevance. Not surprisingly, the more accessible a conceptual metaphor is given a lexical concept, the easier it will be to create an appropriate ad hoc concept and to understand a verbal metaphor. Of course, context also constrains the construction of ad hoc categories. Assume that in the case of “Robert is a bulldozer,” the context, as part of the mutual cognitive environment holding at that moment for the speaker and listener, primes the incorporation of the mind as machine metaphor, such as when Robert or someone else is previously described as a powerful thinker. This conceptual metaphor provides many possible mappings between the source domain machine and the target domain mind. At this point, a relevance-guided sorting process determines which of the many possible source-to-target domain mappings of the machine to mind metaphor may be applied to constructing the ad hoc concept bulldozer* that meets the expectation of optimal relevance for this context. Overall, then, the creation of ad hoc categories during metaphor understanding incorporates people’s


tacit knowledge of conceptual metaphors and the processing constraints imposed by the communicative principle of relevance (Gibbs and Tendahl in press). A similar integrating of conceptual and pragmatic information may shape metonymy processing. For instance, the simple expression “Bush invaded Iraq” may immediately evoke several different conceptual metonymies such as person for nation, person for event, or person for place. Which one of these provides the most optimally relevant, and contextually-appropriate reading will be constrained by a variety of pragmatic factors, including whether the expression was encountered in a news story, a political commentary, or in answer to the question during a political debate “Who’s responsible for the tragedy in Iraq these days?” The ad hoc construction of Bush referring to the person primarily responsible for the United States invading Iraq in the most recent Gulf war is not accomplished by the activation of a single conceptual metonymy, but through pragmatic adjustment regarding several possible conceptual metonymies. Moreover, the conceptual prominence that the metonymic target “Bush” has in a discourse situation is not just due to the activation of some conceptual metonymy, but is more precisely shaped by how the pragmatics at hand supports optimally relevant readings (with these being constrained by the principle that utterance interpretation is guided by a trade-off of maximizing cognitive effects while minimizing cognitive effort (Sperber and Wilson 1995)). For instance, if the expression “Bush invaded Iraq” is given as a reply to the question “Who’s responsible for the tragedy in Iraq these days,” the metonymic reading of “Bush” will highlight President Bush’s individual role in initiating the Iraq war. In a different context, such as when two people are describing the role that various nations had in the Iraq war, the statement “Bush invaded Iraq” will highlight Bush as standing for the United States and perhaps decrease President Bush’s individual culpability in waging the Iraq war. Finally, metonymic expressions provide certain cognitive effects that are also influenced by pragmatics. According to relevance theory, cognitive effects are achieved by one of the following three types: (1) new information provided by a contextual implication, (2) strengthening of an existing assumption, and (3) a contradiction and possible elimination of an existing assumption. Previous research has shown how metaphorical statements, such as “Lawyers are sharks” can be used to strengthen a previous speaker’s negative claims about lawyers, add new information to what a previous speaker said about lawyers, or contradict what another speaker had previously said about lawyers (Gibbs and Tendahl 2006). In each case, our reading of exactly the way that lawyers are sharks differs to some degree. A similar possibility surely exists for understanding metonymic expressions. Hearing “Bush invaded Iraq” may strengthen some idea about Bush, add some new information about Bush, or contradict some current belief about Bush, both as a person and as a metonymic entity that stands-for a place, a nation or an event.

29

30

Raymond Gibbs, Jr.

There are clearly subtle types of interactions that occur between conceptual and pragmatic information when people construct meanings for figurative language. Although it may be the case, as Panther and Thornburg have convincingly demonstrated, that purely pragmatic accounts are incapable of drawing detailed enough metonymic inferences, they are also right that the conceptual approach to metonymy advocated in cognitive linguistics is also insufficient for explaining realistic metonymic readings.

6. Conclusion The study of metonymy has finally come of age in cognitive linguistics, as demonstrated by the work of Klaus-Uwe Panther and others. This research has important connections to psycholinguistic studies of online meaning construction specifically related to metonymy. In general, the empirical research from psychology has not been able to specifically show that people ordinarily access conceptual metonymies as part of their immediate understanding of many metonymic utterances. But the lack of evidence in support of conceptual metonymy is primarily due to the difficulties associated with trying to create experimental tests showing that very abstract mappings, like place for event, are actually recruited during online linguistic processing. Other psycholinguistic evidence, nonetheless, is more directly consistent with cognitive linguistic claims about metonymic language understanding. Studies show that people can quickly draw metonymic inferences during reading and that these inferences make specific targets (i.e., the person in a object for user metonymy such as “The scalpel was sued for malpractice”) more accessible and available in subsequent discourse processing. People’s familiarity with certain metonymic mappings also facilitates their understanding of relevant metonymic utterances, which provides indirect support for the claim that enduring conceptual metonymies may effect online meaning construction. Additional experimental findings are consistent with the claim that metonymy interacts with grammatical structure during immediate processing of certain metonymic expressions. Finally, I have argued that conceptual information strongly interacts with particular pragmatic knowledge when people process metonymic expressions. In this way, conceptual metonymies such as person for nation may constrain the pragmatic interpretation of expressions like “Bush invaded Iraq” by highlighting certain aspects of Bush (i.e., his standing as a symbol for the United States), while diminishing others (i.e., Bush’s individual responsibility for the Iraq war). This possibility is directly in line with cognitive linguistic proposals that pragmatics alone is insufficient to explain immediate meaning construction and needs to be reinforced by


available, and often quite accessible, metonymic mappings, such as those potentially represented by conceptual metonymies. The focus of this chapter on metonymy in online meaning construction serves as a case study of the important influence that cognitive linguistic work has had on experimental psycholinguistics. There remain many questions about how to best experimentally test ideas from cognitive linguistics, and whether some of the putative conceptual schemes discovered by cognitive linguists are indeed psychologically real. But it is also clear that the major advances in the study of metonymy and the novel claim that metonymy is a form of thought and not just a rhetorical device, will have significant implications for psychological theories of how people conceive of certain topics and apply that knowledge during online meaning construction.

References Carston, Robyn. 2002. Thoughts and Utterances: The Pragmatics of Explicit Communication. Oxford: Blackwell. Frisson, Steven and Pickering, Martin. 1999. The processing of metonymy: Evidence from eyemovements. Journal of Experimental Psychology: Learning, Memory and Cognition 25: 1366–1383. Gibbs, Raymond W., Jr. 1990. Comprehending figurative referential descriptions. Journal of Experimental Psychology: Learning, Memory and Cognition 16: 56- 66. Gibbs, Raymond W., Jr. 1992. What do idioms really mean? Journal of Memory and Language 31: 485–506. Gibbs, Raymond W., Jr. 1994. The Poetics of Mind: Figurative Thought, Language, and Understanding. New York: Cambridge University Press. Gibbs, Raymond W., Jr. 1999. Speaking and thinking with metonymy. In Metonymy in Language and Thought, Klaus-Uwe Panther and Günter Radden (eds), 61–76. Amsterdam: John Benjamins. Gibbs, Raymond W., Jr. and O’Brien, Jennifer. 1990. Idioms and mental imagery: The metaphorical motivation for idiomatic meaning. Cognition 36(1): 35–64. Gibbs, Raymond W., Jr., Bogdonovich, Josephine, Sykes, Jeffrey and Barr, Dale J. 1997. Metaphor in idiom comprehension. Journal of Memory and Language 37: 141–54. Gibbs, Raymond, Strom, L., and Spivey-Knowlton, Michael J. 1997. Conceptual metaphor in mental imagery for proverbs. Journal of Mental Imagery 21: 83–110. Gibbs, Raymond W., Jr., Lima, Paula and Francuzo, Edson. 2004. Metaphor is grounded in embodied experience. Journal of Pragmatics 36: 1189- 1210. Gibbs, Raymond W., Jr. 2006. Embodiment and Cognitive Science. New York: Cambridge University Press. Gibbs, Raymond Jr. and Tendahl, Markus. 2006. Cognitive effort and effects in metaphor comprehension: Relevance theory and psycholinguistics. Mind & Language 21: 379–403. Glucksberg, Sam. 2001. Understanding Figurative Language: From Metaphor to Idioms. New York: Oxford University Press. Grice, Herbert Paul. 1989. Studies in the Way of Words. Cambridge: Harvard University Press.

31

32

Raymond Gibbs, Jr.

Markert, Katja and Nissim, Malvina. 2003. Corpus-based metonymy analysis. Metaphor and Symbol 18: 175–88. Panther, Klaus-Uwe and Radden, Günter (eds). 1999. Metonymy in Language and Thought. Amsterdam: John Benjamins. Panther, Klaus-Uwe and Thornburg, Linda (eds) 2003. Metonymy and Pragmatic Inferencing. Amsterdam: John Benjamins. Panther, Klaus-Uwe and Thornburg, Linda. 2004. The Role of Conceptual Metonymy in Meaning Construction. metaphorik.de 6: 91–116. Pfaff, Kerry L., Gibbs, Raymond W., Jr. and Johnson, Michael D. 1997. Metaphor in using and understanding euphemism and dysphemism. Applied Psycholinguistics 18: 59–83. Pickering, Martin and Traxler, Matthew. 1998. Plausibility and recovery from garden paths: An eye-tracking study. Journal of Experimental Psychology: Learning, Memory, and Cognition 24: 940- 961. Sperber, Dan and Wilson, Deirdre. 1995. Relevance: Cognition and Communication (2nd edition). Oxford: Blackwell.

chapter 2

High-level metaphor and metonymy in meaning construction* Francisco José Ruiz de Mendoza Ibáñez and Ricardo Mairal Usón University of La Rioja/National University of Distance Education, Madrid

This chapter addresses the problem of meaning construction in grammar from the point of view of high-level metaphor and metonymy. We argue that meaning construction, in the sense given to this term by Panther (2005), is essentially an inferential activity whether at the level of so-called pragmatic meaning (e.g. implicature derivation, illocutionary meaning) or when dealing with core grammar phenomena, especially conversion processes and constructional alternation, which we explore within the domain of transitivity. In our study we are able to find high-level metaphor and metonymy at the basis of some of these grammatical processes. We also observe that the same principles and conceptual patterns (e.g. metaphor-metonymy interaction and metonymic chaining) that constrain metaphorical and metonymic activity at other levels of analysis are operational at the grammatical level too. Keywords: alternations, coercion, construction, double metonymy, grammatical conversion, high-level metaphor, high-level metonymy, invariance, transitivity

1. Introduction In Cognitive Linguistics metaphor is generally defined as a cognitive mapping (or set of correspondences) across discrete conceptual domains (Lakoff 1993), while metonymy is seen as a domain-internal mapping where one of the domains involved provides a point of access to the other (cf. Kövecses and Radden 1998; see also Barcelona 2002 and Radden 2002 for definitional and demarcation issues). Both processes are central to our understanding of how meaning is constructed in actual language use, since they regulate much of our inferential activity, including pragmatic implications (Panther 2005).

* Financial support for this research has been provided by the DGI, Spanish Ministry of Education and Science, grants HUM2004–05947–C02–01/FILO and HUM2005–02870/FILO. The research has been co-financed through FEDER funds.

34

Francisco José Ruiz de Mendoza Ibáñez and Ricardo Mairal Usón

One of the directions of recent research in Cognitive Linguistics is the connection between metonymy and grammatical description. Some preliminary attempts may be found in work by Kövecses and Radden (1998), Panther and Thornburg (1999, 2000), Ruiz de Mendoza and Pérez (2001). A more ambitious project is under way now in a collective volume edited by Thornburg, Panther, and Barcelona (forthcoming), whose leading article, “Metonymic grammar,” by Ronald Langacker, argues that grammatical links are basically metonymic to the extent that meaning is not compositional and linguistic expressions serve as points of access to extra conceptual material. Thus, a compositional theory cannot predict why the meaning of ‘mosquito net’ (used to protect ourselves from mosquitoes) does not follow the same interpretation pattern as ‘butterfly net’ (used to catch butterflies). Langacker argues that a grammatical relationship between two elements can be established on the basis of the mental access they provide to the entities needed for conceptual integration. However, in our view grammar is not only metonymic in the sense that the relationship between elements of formal structure and meaning is pervasively unpredictable or, as Langacker puts it, indeterminate. As we will show, there is a significant amount of linguistic evidence that many grammatical phenomena are grounded in metonymy and that we need to have a good understanding of how metonymic activity takes place in order to understand their semantic makeup and thereby their communicative potential. What is more, we will argue that metonymy is not the only motivating factor underlying grammatically guided meaning construction. Metaphor also plays a very significant role in grammatical organization and our consequent ability to exploit grammar to construct meaning.

2. Levels of description Understanding the cognitive grounding of grammar is contingent on considering description levels for metaphor and metonymy. This issue has been addressed in detail in Ruiz de Mendoza (2005a), where it is argued that it is possible to find a unified framework for inferential and grammatical description by distinguishing between high and low levels of description, on the one hand, and situational and non-situational cognitive models, on the other. Thus, implicature is seen as the result of metonymic activity on low-level situational models; illocutionary meaning, especially where there is inferential activity, takes place by applying metonymic expansion operations to partial structure from high-level situational models. Interestingly enough, many grammatical processes also seem to have an inferential nature, which calls for an explanation of their semantic import in terms of metonymic operations on high-level (i.e. generic) non-situational cognitive models (tradition-

High-level metaphor and metonymy in meaning construction

al examples of referential metonymy are simply operations on low-level non-situational cognitive models, as in order for customer and ruler for army). For example, think of the meaning implications of the following sentences:

(1) I think I’m getting cold. (2) The door opened. (3) I saw the destruction of the city.

One favored interpretation of sentences like (1) is that the speaker is indirectly asking the addressee to take action to solve the problem he is complaining about. In a theory that admits the existence of high-level (or generic) cognitive models, whether of a situational or a non-situational kind, the so-called “illocutionary value” of (1) can be explained by postulating the activity of a metonymy whereby part of a high-level situational model stands for the whole of it (this kind of model is what Panther and Thornburg (1998) have called an illocutionary scenario). The high-level model is based on the cultural convention that we are expected to help other people when needed. This convention is based on high-level or generic structure derived from many low-level everyday situations that specify the ways in which people make requests or otherwise engage in directive illocutionary activity. Making the addressee aware that we are in need is a way of calling attention to the whole situation in which people are expected to help and consequently counts as a request for help. Now consider example (2). This sentence is generally treated as a case of the inchoative construction, which in English alternates with the causative construction (cf. Someone opened the door). In interpreting (2) the language user is aware that doors do not open without an external cause (a person, an animal, the wind, etc.), so in the appropriate contexts (2) would give rise to more or less specific inferences as to the nature of the controlling entity. Imagine that we have heard a dog growl behind the door and then we see the door move open. In that context, a valid inference would be ‘The dog opened the door.’ This inference is retrieved on the basis of a high-level metonymic process, which we will label process for action (cf. Ruiz de Mendoza and Pérez 2001). Finally, consider sentence (3), where the event of destroying a city is treated as a physical entity. It may be argued that this sentence is an example of the high-level metaphor events are entities. Postulating this metaphor allows us to predict a number of meaning implications: some participant entities may be left implicit thereby being backgrounded (e.g. the agent of the action); the event is seen as an encapsulated piece of experience that can be treated as an instrument (We were threatened with destruction), as an object of perception (We saw their destruction) or of knowledge and understanding (We knew about their destruction); however, since the nominalization preserves the internal structure of the event, it is possible

35

36


to make use of it in terms of time relationships (We were able to put an end to the destruction of the city). The idea that some grammatical processes have a metaphorical explanation is also found in Halliday (1994) under the label grammatical metaphor, although the differences are significant. Besides the obvious case of nominalizations, Halliday includes within this notion such linguistic phenomena as non-literal modality expressions (e.g. I think she knows as a grammatically non-congruent version of She probably knows), and mood (or speech act) expressions (e.g. I wouldn’t do that if I were you as a non-congruent form of Don’t do that!). High-level metaphor results from finding correspondences between high-level propositional cognitive models. Halliday’s grammatical metaphor, in contrast, is a matter of applying linguistic resources from one system to another system. For example, we may express probability in a congruent way by means of the adverb “probably,” which is a modality resource, or by making use of non-congruent hypotactic projection within the logical subcomponent of the ideational system (“I think” introducing “she knows”). We do not follow Halliday’s proposal for two crucial reasons. One is that metaphor is not necessarily a matter of using language non-congruently, as has been demonstrated by the pervasiveness of this phenomenon in language (cf. Lakoff and Johnson 1980, 1999). Instead, metaphor can be understood as a cognitive process that allows us to reason about a conceptual domain in terms of the structure and logic of another domain. The other reason is that what Halliday regards as “non-congruent versions within a system” of “congruent” grammatical options from a different system may be more simply treated as different constructional realizations that express relatively similar meanings. Thus, there is nothing metaphorical in expressing a warning by means of an if-conditional complex clause. In fact, there is no principled reason why using an imperative to issue a warning is more congruent than using a conventional if-conditional construction. From the perspective of grammatical mood, a negative imperative maps onto a command not to do something, but this is not necessarily a warning (cf. Don’t do the ironing today; I’ll do it so you can have some rest). From our discussion, it is evident that our proposal to explain some grammatical processes on the basis of high-level metaphorical and metonymic operations has two clear advantages: (i) it allows us to capture relevant inferences or meaning implications whose actual communicative impact may otherwise be lost from our description; (ii) it allows us to introduce a greater degree of regularity in our descriptive and explanatory apparatus, thus enhancing the predictive power of the model. We now turn our attention to the question of general constraints on metaphor and metonymy since these constraints carry over to cases of high-level metaphor and metonymy, as will be seen in section 5.


3. Constraints on metaphor and metonymy One of the problems with the original formulation of the cognitive theory of metaphor was to account for the way in which a linguistic description may rule out cases of non-metaphor on a principled basis. The Invariance Principle (Lakoff 1993) was originally formulated as a constraint on metaphor. According to this principle, the topological (i.e. image-schematic) structure of the target domain of a metaphoric mapping has to be preserved in a way that is consistent with the topological structure of the source. For example, if we map a tree onto a person, the top of the three will map onto the person’s head, the branches onto the arms, the trunk onto the body, and so on. It may be argued that the Invariance Principle also applies to metonymy. But here, since metonymies are domain-internal mappings, what is preserved is the basic topological configuration underlying the mapping. Thus, the metonymy from bottle to wine (e.g. He has been drinking bottle after bottle) preserves the structural container-contents relationship of the ‘container’ image schema in such a way that since we think of wine as being held inside a bottle, it is possible to make the bottle stand for the wine in the context of drinking (cf. *He wants to drink another cork, which disrupts the basic container-contents relationship). However, the Invariance Principle is limited to cases of metaphor based on image schemas. In order to cover non-topological metaphors, Ruiz de Mendoza (1998) formulated the Extended Invariance Principle, according to which the generic-level structure of the metaphoric target has to be preserved in a way that is consistent with the generic-level structure of the source. The Extended Invariance Principle applies conceptual consistency constraints to all kinds of high-level structure. Thus, a person’s robustness may find an appropriate correlate in the strength of a tree or of any other object, such as the bodywork of car. In the case of metonymy, the principle applies by preserving the generic-structure configuration of domain-internal relationships; e.g. the controlled-controller relationship between ‘bus-driver’ and ‘bus’ is preserved in The buses are on strike, but not in *The wheels are on strike. The Extended Invariance Principle focuses on adequately correlating source and target elements. Thus, in love is a journey, travelers naturally map onto lovers since both are controlling entities that develop a goal-oriented activity, and the love relationship maps onto a vehicle since both the relationship and the vehicle are ways of achieving the intended aims. Note that the source elements are but role characterizations that need to be realized by specific frame elements. It is here where there is room for potential oddities. Think for example of the infelicity of talking about a relatively unimportant difficulty in the love relationship in terms of a serious physical impediment to motion, such as a ravine or a steep hillside. The Extended Invariance Principle is insufficient to rule out such potential oddities. In order to deal

37

38


with this problem, Ruiz de Mendoza and Santibáñez (2003) have formulated the Correlation Principle, which places constraints on metaphoric mappings on the basis of the implicational structure of source and target elements. Thus, for a metaphoric source element to qualify as the counterpart of an element in the target domain, the source element needs to share the relevant implicational structure of the target element. For example, Our relationship crashed right after takeoff is about a relationship that looked strong and promising, but has been abandoned in a dramatic, unwanted way. The source domain could hardly have a vehicle other than an airplane to yield the right meaning effects. This kind of vehicle is powerful and can cover very long distances in very little time, which readily maps onto the idea that the love relationship was expected to grow strong and achieve success in a comparatively short period of time. But this expectation is thwarted. In its application to metonymy, the Correlation Principle is also active in the selection of the most relevant source domain in terms of its potential to afford access to the intended target domain. In The ham sandwich is waiting for his check, the kind of meal ordered is not the only possible way for the waitress to achieve her referential purpose (cf. The hooked nose/The fur coat/The muddy shoes is waiting for his check) but it is a straightforward, convenient way in the fast food restaurant context. Another relevant principle is the Mapping Enforcement Principle (Ruiz de Mendoza, 2005b), which ensures that no item in the target will be discarded from a mapping system if there is a way to find a corresponding item in the source. Consider the expression He gave John a kick. Lakoff (1993) notes that here an action is seen as a transfer of possession, but the person who figuratively receives the kick does not have it afterwards. For Lakoff, the Invariance Principle would seem to constrain the mapping in such a way that the possession element from the transfer schema has to be discarded since there is no corresponding element in the target. The rest of the Source

Target

giver

kicker

receiver

kickee

object

kick

giving

kicking

possession

effects of kicking

Figure 1. An action is a transfer of possession


target elements seem to have perfect matches in the source: the agent is the giver; the patient is the receiver; and kicking is giving. There is a problem with this account, though, since one of the crucial meaning effects of ‘giving a kick’ is that the person who ‘gets the kick’ is affected by it, i.e. the person does not have the kick but rather the effects of the kick. A solution to this problem is to postulate the existence of a metonymic mapping from ‘kicking’ to the ‘effects of kicking’ (i.e. action for effect of the action) in the target. This metonymy, which is a consequence of the Mapping Enforcement Principle, enables us to preserve the possession element in the metaphoric source (see Figure 1).

4. Metonymic chains The Mapping Enforcement Principle is active not only in the metonymic development of some metaphors but also in the production of double metonymies (cf. Ruiz de Mendoza and Díez, 2002) also called metonymic chains (Barcelona, 2002: 266). Consider these examples:

(4) I love Proust (‘Proust’s work’) (5) Proust is on the top shelf (‘a book containing Proust’s work’)

In (4) we have a clear case of author for work. In (5), however, even though Proust’s work is involved, this single mapping is insufficient to account for the full semantic import of the expression. We need a second metonymic mapping from Proust’s work to the medium in which it is presented, thus yielding the double metonymy author for work for medium. Since metonymy is based on domain-inclusion relationships where the source may be a subdomain of the target or the target a subdomain of the source (cf. Ruiz de Mendoza, 2000), metonymy consists basically of domain expansion and reduction operations. This allows us to predict four double metonymy types (see Figures 2a–d, p. 40): a. Double domain reduction: Wall Street is in panic. b. Double domain expansion: His sister heads the policy unit. c. Domain reduction and domain expansion: Proust is on the top shelf. d. Domain expansion and domain reduction: He has too much lip.

5. Metaphor, metonymy, and grammar Our discussion of metaphor and metonymy has provided us with a number of analytical tools for understanding meaning construction processes. These tools are re-

39

40


place

action of leading institution

leader/agent people

Figure 2a. place for institution for people

head

Figure 2b. head for leader for leading

action

author work medium/ format

Figure 2c. author for work for medium

instrument

ability

Figure 2b. instrument for action for ability

lated to two crucial areas on enquiry in cognitive modeling: (i) the existence of a high-level of description; (ii) the existence of principles that constrain metaphoric and metonymic activity. The following sections study the way in which these analytical tools bear on the explanation of some instances of how meaning is constructed in grammar on the basis of high-level inferential activity. For reasons of space limitations we will restrict our discussion to the domain of transitivity.

5.1 The action frame and transitivity Following Halliday (1994), we mean by transitivity the area of grammar concerned with types of processes and their relationships with the central participants, attributes and circumstances. We relate this area of grammar to a high-level propositional idealized cognitive model, which we will refer to as the action frame. The structure of this cognitive model has been the object of study of linguistic theories


that take into consideration semantic functions (or roles) (e.g. agent, object, instrument, etc.) and related semantic parameters such as number of participants, kinesis, aspect, punctuality, volition, affectedness of object, among others (cf. Hopper and Thompson 1980). One important assumption of such theories is that, at some level of description, a linguistic expression (to the extent that it sets up relationships between predicates and their arguments) designates a state of affairs that can be assessed in terms of parameters like dynamism, control, telicity, duration, and others. For example, in Dik’s Functional Grammar (Dik 1997), which makes use of some of Vendler’s seminal Aktionsart distinctions (cf. Vendler 1967), actions are defined as dynamic controlled events, to be contrasted with processes, which are dynamic but uncontrolled. Actions are subdivided into activities (non-telic; e.g. John was reading) and accomplishments (telic; e.g. John ran the marathon). Processes may be further subdivided into cases of dynamism (non-telic; e.g. The clock was ticking) or of change (telic; e.g. The apple fell from the tree). However, this kind of description has two limitations. First, it focuses on a classification of modes of action arising from predicate-argument relationships, thus ignoring other complementary perspectives like the intrinsic nature of actors and objects. In addition, in reducing the distinction between actions and processes to just a matter of a mode of action feature (i.e. control), it misses other points of connection and differences between the two event types. We shall deal with each problem in turn. Actions typically have objects that may be of different kinds, depending on the type of action. Thus, in We all ran the marathon, the marathon is the result of a telic action; in The sheriff killed the gunman, the gunman is an affected object; in John spent hours listening to the same music, the music is the experienced object of a controlled activity; in The boss scolded Jim, a controlled accomplishment, Jim is the receiver (and the goal) of the boss’s remarks. Each object type relates to a different mode of controlled action: resultative, effectual, experiential, and communicative. It is thus possible to classify action predications not only from the point of view of telicity, as Dik does, but also by taking into account the nature of the object of the action (see Table 1). Using this additional criterion to classify modes of action has important consequences for understanding high-level metaphoric activity in grammar, as will be evidenced in the next section. Table 1. Object types within two ‘mode of action’ categories Resultative Effectual Experiential Communicative

Activity

Accomplishment

John was painting portraits In Baghdad some groups will kill people for years He was listening to music The boss kept yelling at him

John painted a portrait The sheriff killed the gunman in a split second He listened to our song The boss scolded him twice

41

42


The second problem concerns the difference between actions and processes. It is true that the control element is crucial in drawing the distinction. Thus, because of the control component actions may have beneficiaries, instruments, and intended results or unintended outcomes. Processes do not typically have beneficiaries (e.g. *The wind blows for me; *The tree fell down for me) or instruments e.g. (*The tree fell down with an axe), and a process may have a natural outcome (e.g. if wood catches fire it burns to ashes). According to this description, processes may be regarded as subdomains of an action frame where the “resultative” component is parametrized differently. However, not all kinds of process are subdomains of the action frame. A process can be regarded as part of an action frame only to the extent that it is possible to retrieve an implicit agent (e.g. ‘dying’ may simply happen or it may be a caused event, which is not necessarily the case with sleeping or dreaming). This feature of action frames makes it possible to create metonymic shifts between actions and processes.

5.2 High-level metaphor and transitivity Consider now the following sentences:

(6) Peter kicked John out of the office. (7) Peter laughed John out of the office.

These sentences illustrate the caused-motion construction (cf. Goldberg 1995). The construction conflates the roles of ‘affected object’ and ‘actor’ into one element of structure (“John” in the examples). It also conflates into one single predicate (‘kick,’ ‘laugh’) two predicate values: causing motion and manner of causing motion. However, (7) differs from (6) in that ‘laugh’ has undergone subcategorial conversion from a verb with a prepositional complement (“laugh at someone”) to a purely transitive verb (“laugh someone”). In our view, the conversion process is licensed by a high-level metaphor whereby we partially map one form of transitive structure onto another form of transitive structure. In this metaphor, the source has an ‘effector’ and an ‘affected object’ that map respectively onto a ‘doer’ and an ‘experiencer’ in the target. As a result of this mapping between different ways of construing the ‘actor-goal’ transitive relationship, we understand an action that has no direct effect on the goal participant as if it did have such a direct effect. We may label this metaphor, following our previous classification of object types within modes of action, experiential action is effectual action. . Lack of space prevents a detailed discussion of semantic functions and their labels in connection to the action frame. Note, however, that actors in effectual actions are different from actors in other kinds of action. We refer to the former as “effectors” and to the latter as “doers”.


The reason why this mapping takes place is a consequence of the constraints imposed by the caused-motion construction, which is operational on the basis of effector-affected object relationships, as in (6), but not of doer-experiencer relationships. Michaelis (2003) has discussed a host of cases of lexical adaptation to constructional requirements in terms of the notion of coercion and the related Override Principle, according to which the meaning of a lexical item conforms to the meaning of the structure in which it is embedded. In the case of ‘laugh’ above, this principle predicts that the arguments of the verb have to be re-construed according to semantic coherence constraints. However, this explanation misses the important fact that the adaptation process is grounded in a high-level metaphor. Thus, the full meaning impact of (7) may not be understood independently of the metaphorical mapping: we understand one form of being affected by an action (as an experiencer) in terms of another form of being affected by an action (as an affected object). This mapping is licensed by the combined activity of the Extended Invariance and Correlation principles. In fact, the possibility of correlating the affected object and the experiencer is enhanced by the fact that both generic-level elements share relevant implicational structure (both are goals of an actor’s action). However, we do not map all possible elements of the transitive structure. The role ‘instrument’ from the source is discarded in accordance with the Extended Invariance Principle since the target domain (the goal-oriented activity of laughing at someone) only requires correspondences for the actor/doer and the goal/experiencer of the action but there is no instrument (cf. *John laughed at Peter with his mouth). Let us now take another case of high-level metaphor within the area of transitivity:

(8) He talked me into it.

Here ‘talk someone (into)’ results from a metaphorically-grounded subcategorial conversion process whereby a goal-directed intransitive expression (‘talk to someone’) is transitivized. In the metaphor (communicative action is effectual action) the receiver of the message is seen as if directly affected by the action of talking rather than as the goal of the message. Noting this correlation is essential to understanding all the meaning implications of the sentence. As with (7), the metaphor in (8) is sensitive to all the principles and constraints that we have identified for metaphor. All relevant target elements, through the application of the Extended Invariance Principle, find corresponding elements in the source: He talked me into it with persuasive words (instrument) just for John’s sake (beneficiary). The receiver-affected object mapping is licensed by the Correlation Principle since both elements are goals of an actor’s action. Finally, in (8) caused motion is metaphorical and maps onto the notion of persuasion (being persuaded is being introduced into a container). This metaphorical process is unrelated to

43

44


the underlying high-level mapping that correlates the receiver and affected object elements. Another possible high-level metaphorical mapping, a (non-effectual) activity is an effectual accomplishment, is illustrated by examples (9) and (10): (9) He drank himself into a stupor daily. (10) She collapsed and laughed herself out to silence. These sentences are examples of the resultative construction. As with caused motion, this construction often forces subcategorial conversion of intransitive and pseudo-transitive verbs into transitive predicates. In (9) the originally intransitive ‘drink’ is understood in terms of a transitive structure of the reflexive actor- object kind, in order to make it compatible with the caused-motion structure underlying the resultative construction. In (10), “laugh (out)” is transitivized differently from (7) since ‘herself ’ is not treated as an affected object but as the object of a change. In resultatives the caused element is the result of a process, whether seen in figurative terms or not. In caused-motion constructions, the caused element is always a process (of going to a different place). Note that ‘talk someone into’ is not a resultative expression, since it focuses on the process of being persuaded rather than on the result of the process. However, ‘drank himself into a stupor’ and ‘laughed herself out to silence’ focus on the resultant state, which is explicitly invoked by the linguistic expression and cues the constructional choice. To conclude this section, consider a more complex example of high-level metaphor, metaphorical communicative action is effectual action, within the domain of transitivity: (11) The speaker addressed a few introductory remarks to the audience. (12) The speaker addressed the audience with a few introductory remarks. As is well known from the literature on constructional alternations, (11) and (12) are alternative forms of construing the same event. Both are metaphorical, each in a different way. While (11) is a case of the low-level communicating is giving metaphor, (12) departs from the construal in (11) to make use of high-level metaphor communicative action is effectual action. The metaphor in (11) is a cultural model about the transmission of meaning that has been explored by cognitive linguists in terms of the so-called conduit metaphor (according to which ideas are objects, words are containers for these objects, and communication is sending and receiving these containers through a conduit; Reddy 1979). As a cultural model, although metaphorical, it may be taken as the default version of how we understand communication. A sentence like (12) is a departure from the default model, which allows us to think of ‘addressed the audience’ as the result of a conversion process in very much the same way as has been postulated for


examples (7)–(10). Thus, sentence (12) makes use of a purely transitive construction where the audience, which is the goal (seen as receiver) of a communicative action, is metaphorically presented as the affected object of an instrumental action. This has obvious meaning effects: while in (11) the focus is on the members of the audience receiving the remarks, in (12) the focus is upon the effect of the remarks on the audience. There are other alternations within the action frame which we have not explored here whose activity is internal to the effectual action type and do not seem to involve high-level metaphor. Compare, for example, John hit the wall with a stick and John hit a stick against the wall. The second sentence provides us with an alternate construal of the canonical event where someone uses a stick to hit a wall. In the alternate construal the true object of the action (the wall, an affected object) is seen as a targeted location, while the instrument is seen as an object that is moved into contact with the target location. The interpretation of the second sentence may not involve metaphor since “the wall” is actually both the object of the action and also the place where the stick hits.

5.3 High-level metonymy and transitivity There are verbs that can take part in different constructions thus giving rise to constructional alternations, such as the causative-inchoative and the middle alternations (Levin 1993). We now explore the two alternations and argue that they are grounded in high-level metonymy, since they exploit different aspects of the ‘action’ high-level propositional model (or ‘action’ frame). However, high-level metaphor plays no role in the types of subcategorial conversion involved in these constructional alternations. The reason for this is to be found in the fact that these processes are domain-internal. Unlike the case of the conversion of ‘laugh at’ into ‘laugh,’ we are not dealing with two different kinds of transitivity relationship here, but with different (highlighted) aspects of one kind of transitivity. Compare the following sentences, which seem to share some properties while keeping clear differences:

(13) The door closed well thanks to her instructions. (14) This bread cuts easily/well [cf. It is easy/*well to cut this bread]. (15) This soap powder washes whiter. bread cuts. (16) *This *This soap powder washes. (17)

A sentence like This bread cuts easily is usually regarded as a case of the middle construction that can be paraphrased in terms of a preparatory subject structure where the evaluative adverb is converted into a predicative adjective with the anticipat-

45

46


ACTION

PROCESS

Figure 3. process for action

ACTION

PROCESS

RESULT

Figure 4. process for action for result

ed predication under its scope (i.e. It is easy to cut this bread). This kind of analysis would seem to call for an underlying metonymic shift of the process for action kind (Figure 3). However, in The door closed the manner adverb expressing assessment is optional, while in (14) it is compulsory. This suggests that we have completely different processes within the action frame. In order to capture the full range of meaning implications of This bread cuts easily in compliance with the Mapping Enforcement Principle we need to postulate a double metonymy. process for action for result (Figure 4), is based on our previous construal of an action (cutting bread) as if it were a process (the bread “cuts”) that is assessed in terms of the inherent difficulty in achieving a result (the bread gets cut without difficulty). In this complex metonymic shift, the process of getting cut is linguistically presented as central, while the result remains in the domain of implicit meaning. This situation differs slightly from one in which we use the adverb “well” in (14), as revealed by the impossibility of making a paraphrase in terms of the preparatory subject construction (cf. *It is well to cut this bread). The reason for this lies in the fact that this adverb explicitly focuses on the result of the action, while “easily” focuses on the process. The metonymic process is essentially the same, but each adverb gives prominence to a different aspect of the underlying action, as evidenced by the preparatory “it” test. A sentence like This soap powder washes whiter, on the other hand, also exploits the ‘result’ part of the process for action for result metonymy, just like This bread cuts well, but in a different way. While (14) bases its assessment on the properties of the object of the action, in (15) the focus is on the instrument of the action. Note that, whichever the case may be, we assess neither the object nor the instrument but the outcome of acting on the object or of making use of the instrument to perform the action. This kind of analysis has the advantage of emphasizing


the characteristics that the so-called ‘middle’ and ‘characteristic property of instrument’ constructions have in common. Finally, (16) and (17) are odd because, in the absence of any overt indication that process for action for result is to be invoked, we tend to interpret the predicate in terms of process for action; this interpretation runs counter to intuition since, while it is easy to envisage a door opening without focusing on the agent (e.g. when the wind blows), it is difficult to think of bread or soap powder performing any action by themselves.

6. Conclusion In this chapter we have given evidence that lexical and constructional representations interact in predictable ways and that many of the resulting processes are grounded in high-level metaphor and metonymy, i.e. metaphoric and metonymic operations performed on high-level (or generic) non-situational cognitive models such as action, process, cause, perception, etc. We have discussed some of the constraints that apply to metaphor and metonymy and observed the way in which such constraints allow us to account for a relevant part of the meaning implications of a number of grammatical phenomena, including several cases of subcategorial conversion and constructional alternation. In our discussion it has become evident that high-level metaphor and metonymy regulate the way constructional coercion, as described by Michaelis, overrides lexical meaning in the part of meaning construction that pertains to the adaptation of verbal predicates to generic constructions, such as caused-motion. Such meaning implications are an essential part of meaning construction processes, in the sense given to this term by Panther (2005).

References Barcelona, Antonio. 2002. Clarifying and applying the notions of metaphor and metonymy within Cognitive Linguistics: An update. In Metaphor and Metonymy in Comparison and Contrast, René Dirven and Ralf Pörings (eds.), 207- 277. Berlin: Mouton de Gruyter. Dik, Simon C. 1997. The Theory of Functional Grammar. Part 1: The Structure of the Clause (Edited by Kees Hengeveld). Berlin: Mouton de Gruyter. Goldberg, Adele. 1995. A Construction Grammar Approach to Argument Structure. Chicago IL: University of Chicago Press. Halliday, Michael A.K. 1994. An Introduction to Functional Grammar (2nd. ed.). London: Edward Arnold. Hopper, Paul J. and Thompson, Sandra. 1980. Transitivity in grammar and discourse. Language 56(2): 251–99.

47

48


Kövecses, Zoltán and Radden, Günter. 1998. Metonymy: Developing a cognitive linguistic view. Cognitive Linguistics 9: 37–77. Lakoff, George and Johnson, Mark. 1980. Metaphors We Live By. Chicago IL: University of Chicago Press. Lakoff, George and Johnson, Mark. 1999. Philosophy in the Flesh. New York NY: Basic Books. Lakoff, George. 1993. The contemporary theory of metaphor. In Metaphor and Thought (2nd. ed.), Andrew Ortony (ed.), 202–51. Cambridge: Cambridge University Press. Levin, Beth. 1993. English Verb Classes and Alternations: A Preliminary Investigation. Chicago IL: University of Chicago Press. Michaelis, Laura. 2003. Word meaning, sentence meaning, and syntactic meaning. In Cognitive Approaches to Lexical Semantics, Hubert Cuykens, René Dirven and John R. Taylor (eds), 93–122. Berlin: Mouton de Gruyter. Panther, Klaus-Uwe and Thornburg, Linda. 1998. A cognitive approach to inferencing in conversation. Journal of Pragmatics 30: 755–69. Panther, Klaus-Uwe and Thornburg, Linda. 1999 The potentiality for actuality metonymy in English and Hungarian. In Metonymy in Language and Thought, Klaus-Uwe Panther and Günter Radden (eds.), 333–57. Amsterdam: John Benjamins. Panther, Klaus-Uwe and Thornburg, Linda. 2000. The effect for cause metonymy in English grammar. In Metaphor and Metonymy at the Crossroads, Antonio Barcelona (ed.), 215–32. Berlin: Mouton de Gruyter. Panther, Klaus-Uwe. 2005 The role of conceptual metonymy in meaning construction. In Cognitive Linguistics. Internal Dynamics and Interdisciplinary Interaction, Francisco Ruiz de Mendoza and M. Sandra Peña (eds.), 353–86. Berlin: Mouton de Gruyter. Radden, Günter. 2002. How metonymic are metaphors? In Metaphor and Metonymy in Comparison and Contrast. René Dirven and Ralf Pörings (eds.), 407–34. Berlin: Mouton de Gruyter. Reddy, Michael J. 1979. The conduit metaphor: A case of frame conflict in our language about language. In Metaphor and Thought, Andrew Ortony (ed.), 284–97. Cambridge: Cambridge University Press. Ruiz de Mendoza Ibáñez, Francisco J. and Díez Velasco, Olga. 2002. Patterns of conceptual interaction. In Metaphor and Metonymy in Comparison and Contrast, René Dirven and Ralf Pörings (eds.), 489–532. Berlin: Mouton de Gruyter. Ruiz de Mendoza Ibáñez, Francisco J. and Pérez Hernández, Lorena. 2001. Metonymy and the grammar: Motivation, constraints, and interaction. Language and Communication 21: 321– 57. Ruiz de Mendoza Ibáñez, Francisco J. and Santibáñez Sáenz, Francisco. 2003. Content and formal cognitive operations in construing meaning. Italian Journal of Linguistics 15(2): 293– 320. Ruiz de Mendoza Ibáñez, Francisco J. 1998. On the nature of blending as a cognitive phenomenon. Journal of Pragmatics 30(3): 259–74. Ruiz de Mendoza Ibáñez, Francisco J. 2000. The role of mappings and domains in understanding metonymy. In Metaphor and Metonymy at the Crossroads, Antonio Barcelona (ed.), 109– 32. Berlin: Mouton de Gruyter. Ruiz de Mendoza Ibáñez, Francisco J. 2005a. High level cognitive models: In search of a unified framework for inferential and grammatical behavior. In Perspectives on Metonymy, Krzysztof Kosecki (ed.). Frankfurt: Peter Lang.


Ruiz de Mendoza Ibáñez, Francisco J. 2005b. Construing meaning through conceptual mappings. In Lengua y Sociedad: Aportaciones Recientes en Lingüística Cognitiva, Lingüística del Corpus, Lenguajes de Especialidad y Lenguas en Contacto, Pedro Fuertes (ed.), 19–38. Valladolid: University of Valladolid. Thornburg, Linda, Panther, Klaus-Uwe and Barcelona, Antonio (eds.). 2006. Metonymy and Metaphor in Grammar. Amsterdam: John Benjamins. Vendler, Zeno. 1967. Linguistics in Philosophy. Ithaca NY: Cornell University Press.

49

chapter 3

The role of metonymy in meaning construction at discourse level A case study* Antonio Barcelona University of Murcia

The process of meaning construction, as understood by Panther and Thornburg (Panther 2005), namely the determination of both explicit and implicit meaning, is to a large extent guided or facilitated by conceptual metonymy. The previously published research of the author of this chapter on the function of metonymy in guiding text-level implicatures has been mostly confined to the study of brief written dialogues. The present chapter reports on a case study dealing with this function of metonymy in a relatively extended portion of a narrative text. The findings in the study confirm the hypothesis that metonymy is also the main conceptual mechanism guiding the implicatures invited by this type of text. The study also shows that the metonymies guiding the implicatures respond to a finite set of generic types. Keywords: cognitive linguistics, conceptual metonymy, discourse-pragmatic inferencing, meaning construction, text-level implicature

1. Introduction Metonymy has been shown to be ubiquitous in language and cognition (Barcelona 2002b; Radden 2005). One of the main areas of language where metonymy is particularly relevant as a “guiding” or “triggering” factor is what has recently been referred to as meaning construction . This notion of meaning construction (Panther 2005: 353, reflecting work by Panther and Thornburg) denotes the pragmatic reasoning processes which lead to the determination of explicit pragmatic meaning (i.e. ‘explicatures’; see Sperber and Wilson 1995: 182) and implicit pragmatic meaning (i.e. conversational implicatures). In this chapter I will present some selected parts of a case study devoted to the investigation of the role of conceptual metonymy in an aspect of meaning construc * The research reported on in this chapter has been funded in part by a grant awarded by the Spanish Ministry of Science and Technology (McyT) to research project no. BFF2003–07300 (Project leader: Antonio Barcelona).

52

Antonio Barcelona

tion, namely the derivation of text-level implicatures. This case study is a small part of a much larger, ongoing research project investigating the multilevel operation of metonymy in authentic texts which will include more case studies and analyses (initial results in Barcelona 2005; final results to appear in Barcelona, in preparation). What are here called ‘text-level implicatures’ are implicatures that are instrumental in the reader’s apprehension of the pragmatic connections across several sentences in the text (a written text in this case study) or in the comprehension of the text as a whole, not just of one particular sentence or a part thereof. Some examples of text-level implicatures are those which lead the hearer or reader to identify in the preceding text the antecedent of an anaphorical expression, or those enabling the hearer/reader to identify or hypothesize the purpose of the whole text or text portion, or to anticipate the sort of information provided by the immediately following co-text. Therefore, those implicatures that do not seem to be relevant for the appreciation of text-internal cross-sentence connections or that do not seem to be invited by the text as a whole are not considered text-level implicatures and are not included in the case study. The research hypothesis in the case study is that all or most of the text-level implicatures invited by the text portion chosen for the study will be “guided” or “facilitated,” together with contextual factors and general pragmatic principles, by one or more conceptual metonymies. The hypothesis should not be interpreted as implying that metonymy is the only or even the most important factor triggering the derivation of implicatures. As stated above, metonymy acts in conjunction with contextual factors and with general pragmatic principles (among them, the Relevance Principle and Grice’s maxims). This chapter is organized as follows: Section 2 includes the cognitive linguistic notion of metonymy that I adhere to and briefly discusses the relation between metonymy and pragmatic inferencing; Section 3 is devoted to the discussion of the relevant parts of the case study; and Section 4 briefly presents the conclusions.

2. Metonymy Cognitive linguists regard metonymy as a cognitive mechanism (Lakoff and Johnson 1980, 1999; Lakoff 1987), not merely as a figure of speech, as claimed by traditional rhetoric, nor as a mere “contextual effect,” as claimed by relevance theorists (Papafragou 1996). However, there is not a uniform notion of metonymy in modern linguistics, not even within cognitive linguistics (the general theoretical framework which I follow), and this is why it is necessary to clarify this notion. Some cognitive linguists (e.g. Croft 1993) seem to require every metonymy to be connected

Metonymy in discourse-level meaning construction

to an act of reference, whereas for most others (e.g. Lakoff and Johnson 1980), metonymy is primarily referential, which implies that there can also be non-referential metonymies. Most explicit or implicit definitions of metonymy regard it as a mapping within the same domain (Lakoff and Turner 1989). Some others (Langacker 1993; Kövecses and Radden 1998) instead draw attention to the fact that the source provides “mental access to” or “activates” the target. As we can see, the term ‘metonymy’ is used in cognitive linguistics to cover some very different phenomena, but all cognitive linguists agree on stressing its primarily conceptual nature. It would take a long article to discuss at length the particular variants of this basic conceptual approach to metonymy that are found in the cognitive linguistic literature (see Barcelona 2003a). Given the limitation of space, I will only present and briefly discuss here my own cognitive-linguistic notion of metonymy – schematic metonymy –, which is the type of metonymy exhibiting the properties that, in my view, must be shared by every metonymy: Metonymy is an asymmetrical mapping of a conceptual domain, the source, onto another domain, the target. Source and target are in the same functional domain and are linked by a pragmatic function, so that the target is mentally activated. (Barcelona 2001, 2002a: 246)

The first point to note is that metonymy does not have to be referential. Examples of non-referential metonymies have often been noted in the literature. The term mental activation means that the source is a reference point providing mental access to the target (Langacker 1993; Kövecses and Radden 1998). Mapping refers to the fact that the source domain is connected to the target domain by imposing a perspective on it, not by projecting its structure onto it, as in metaphor. The mapping in metonymy is unidirectional and asymmetrical, whereas the one in metaphor is unidirectional and symmetrical. By “symmetrical” I mean that each element in the source plays a role that is structurally equivalent to the one played by its counterpart in the target. Functional domain (Barcelona n.d., 2002a, 2003a) is equivalent to Fillmore’s (1985) term frame and to Lakoff ’s (1987) term icm (Idealized Cognitive Model). . As Kövecses and Radden (1998) have pointed out, we often find reverse metonymies; but this does not mean that in a given metonymy the mapping proceeds simultaneously in both directions. . In the sentence Picasso is not easy to understand, picasso’s artistic work is a metonymic target, and its activation is carried out from the source picasso, in his role as artist, with the result that the hearer/reader is invited to conceptualize this artistic work primarily as the outcome of Picasso’s artistic genius — as an extension of his personality — other aspects of this work being backgrounded (more details in Barcelona 2003a). But the conceptual structure of the source, picasso, is not systematically and symmetrically projected via this metonymy onto that of the target (as in conceptual metaphor) so that, for example, the artist’s head would correspond to an aspect of his work, his brain to another aspect, etc.

53

54

Antonio Barcelona

As regards the term pragmatic function, it designates a fundamental property of metonymy, namely the fact that the source maps onto, and activates, the target in virtue of the experiential (hence pragmatic) link between the roles each of them performs in the same functional domain. This is why Fauconnier (1997: 11) regards metonymy as a “pragmatic function mapping.” A pragmatic function (term borrowed from Fauconnier 1997) is a strong built-in experiential connection between roles in a frame or icm (cause-effect, author-work, agent-action, etc.). In Barcelona (2003a), a set of additional specific definitions (which, for lack of space, cannot be discussed here) are proposed for the other general kinds of metonymy represented by each of the various different phenomena which are covered by the above schematic definition. This is still a very broad definition of metonymy that would cover prototypical, or at least other undoubted instances of metonymy, as well as other less central, even “marginal,” instances of it. The metonymies operating in the area of meaning construction investigated in the present chapter (textlevel implicature) are all, however, undoubted, uncontroversial instances of metonymy, as will be evident from the ensuing discussion. The role of metonymy in discourse-pragmatic inferencing has already been studied by a number of cognitive linguists, though much more still has to be done. Lakoff ’s (1987: 78–9) discussion of the role of metonymy in certain conversational conventions of Ojibwa and English was one of the earliest contributions in this direction. Gibbs (1994: Chapter 7 and this volume) reviews the psychological reality of the role of metonymy in a number of inferential phenomena such as conceptual anaphors, conceptual tautologies, the mental reconstruction of texts previously read or heard, indirect speech acts, interpretation and reasoning in terms of eponymous verb phrases, film, drama and art conventions and techniques, even literary styles, like that of most nineteenth-century historians, which, according to him, can be defined as metonymic. Klaus Panther and Linda Thornburg are the two cognitive linguists that have devoted the most attention in the past decade to the role of metonymy in various types of pragmatic inferencing. Apart from their important papers on the issue (e.g. Panther and Thornburg 1998; Thornburg and Panther 1997), they have recently edited an important collection of essays devoted to this topic (Panther and Thornburg 2003). In that volume, the essay by Ruiz de Mendoza and Pérez Hernández (2003) argues that metonymic mappings are activated in explicature derivation, and the one by Barcelona (2003b; see also Barcelona 2002b and 2005) investigates the pervasive operation of metonymy in the triggering of implicatures. . As claimed in Barcelona (2002a, 2003a), metonymy is a prototype category in which peripheral and non-prototypical instances occupy various positions on a continuum linking them to prototypical instances. See Peirsman and Geeraerts (2006), for a different prototype view of metonymy.


3. Discussion of some relevant parts of the case study 3.1 Goal and methodology As anticipated in the introduction, I present below some selected parts of a case study investigating the role of conceptual metonymy in one aspect of meaning construction, namely the derivation of text-level implicatures throughout a relatively long portion of a text. In a few recent papers, I reported on studies where I had found that metonymy regulated all the implicatures required for the comprehension of brief written renderings of conversational exchanges consisting of two brief, normally one-sentence utterances (see the studies reported in Barcelona 2003b, 2002b, 2005). But a natural question was whether this metonymic guidance of implicature is only restricted to brief written dialogues or whether it also occurs in other types of written discourse, especially in longer texts. In the present chapter, the text-level implicatures invited by the various sentences in a relatively extended piece of discourse (namely, a six-sentence paragraph of a long written narrative text) are examined carefully with the purpose of checking the correctness of the research hypothesis mentioned in Section 1, namely, that all or most of the text-level implicatures invited by this longer piece of discourse can also be claimed to be “guided,” together with contextual factors and general pragmatic principles, by one or more conceptual metonymies. Given space limitation, I can only present a detailed discussion of the metonymic guidance or facilitation of some of the implicatures invited by the portion selected for the analysis (see below). The methodology applied in the investigation consisted in the following steps: 1. First a text was chosen at random from a number of texts gathered by a colleague in my research group (http://www.um.es/~lincoing) for a different purpose (the study of spatial language). This particular text had not yet been analyzed by me or by anyone else for that nor for any other purpose. This choice was made to avoid the bias that would result from choosing a text already used previously by me or by someone else to investigate metonymy. 2. Within the text, I arbitrarily chose, before even reading it, the very first paragraph as the portion to be subjected to the study. Again, the aim was to avoid selecting a text portion already known to be rich in metonymy. . Therefore, contrary to the views expressed by one of my anonymous reviewers, this chapter is a cognitive-linguistic exercise, and not simply an essay in text linguistics, because its main thrust is the identification and characterization not just of the inferences invited by the text, but also, and fundamentally, of the conceptual metonymies guiding these inferences. The study of conceptual metonymyis a fundamental aspect of cognitive linguistics, since conceptual metonymy is here regarded as a basic cognitive mechanism.

55

56

Antonio Barcelona

3. I then read that portion, the rest of the text and the contextual information on the purpose of the text (namely that it is a hiking report intended as a guide for prospective hikers), which is provided in the website from which it had been downloaded. 4. I made a list of the text-level implicatures that seemed to me to be invited by the portion. 5. Following a suggestion by an anonymous reviewer, I then conducted a small informal experiment in which I asked a group of seven native speakers of English to confirm or reject these inferences, in order to leave out those inferences that may be due to the fact that I am not a native speaker of English, or to other highly subjective factors. The purpose was to examine only implicatures that most native speakers of English reading the same text might have drawn, too. The experiment is included in the second section of the appendix. 6. Finally, I analyzed the implicatures selected in order to find out whether or not they could have been guided by one or more metonymies (in conjunction with general pragmatic principles and contextual factors). The presentation of some of the results of this analysis constitutes the bulk of this section (sub-section 3.3).

3.2 Text under analysis As stated above, the case study is concerned with the analysis of the first paragraph of an authentic narrative-descriptive text randomly chosen from the web (http:// hikingincolorado.org/beth.html). In order to facilitate later cross-reference, I have added a number in square brackets to each of the sentences in the first paragraph, which is reproduced below with no further additions or deletions on my part: [1] If you have ever driven west on Interstate 70 from Denver to the Continental Divide, you have seen Mount Bethel. [2] Several miles (???) before approaching the exit for Hwy. 6 which leads to Loveland Pass you will see a pyramid shaped peak just to the north of the Interstate. [3] This roadway icon stands out like a sore thumb (cliché ?) due to it’s prominence and near perfect triangular shape. [4] As you pass the base of this peak on the highway you will notice two very steep avalanche chutes which lead directly to the summit. [5] This is the route I chose for my ascent. [6] I pulled off the highway to begin my hike from where it looked as though there once may have been a parking area at the base of the western-most avalanche chute and well off the highway right-of-way.

The complete text is reproduced in the Appendix exactly as it appears on the web. The text initiated by this paragraph is a report of a trip to a Colorado mountain with the purpose of hiking on it. The text forms part of a series of reports on . However, no presumption of the existence of a completely predictable set of implicatures derivable from one and the same text portion underlies this experiment (see Brown and Yule 1983: 266 against such a presumption). It is simply meant to avoid taking into account those inferences that none or only very few native speakers would draw.


similar trips to other Colorado mountains, available at http://hikingincolorado.org. This series is intended as a source of information for prospective hikers in the area. Therefore, the reader of this text can expect it to be about Mt Bethel (which figures prominently in the title of the report), to situate geographically this mountain and at least the beginning of the hiking route, to provide a brief description of the shape and general appearance of the mountain and, of course, to provide a story of the arrival and the subsequent hiking experience of the narrator. In this trip report, the mountain and the initial point of the hiking route are situated geographically for the reader throughout the first four sentences in the first paragraph. This is done indirectly, in the form of a brief set of arrival instructions which guide the reader on a mental highway journey to the hiking area. That is, sentence 1 situates the reader in a mental westbound journey on Interstate 70 from Denver to the Rocky Mountains, which continues through sentences 2, 3 and 4 of the first paragraph. The rest of the report, from sentence (5) onwards, is a direct narrative of the actual hiking experience combined with the description, when relevant, of several features of the area.

3.3 The implicatures and the metonymies guiding them The text-level implicatures drawn on the basis of my own reading are listed below, enclosed in double quotation marks. They are grouped in terms of the sentences that (I believe) invite their derivation. The results of the informal experiment referred to in Section 3.1 are presented next to each implicature by indicating its confirmation ratio (e.g. a 6:1 ratio means that six subjects confirmed the implicature and that one did not confirm it; a 7:0 ratio means that all subjects confirmed the implicature). Only those implicatures confirmed by at least six native speakers or implicit in another implicature accepted by six native speakers were selected for analysis. This eliminates six of the implicatures that I had drawn on the basis of my own reading of the passage; they are also listed below. Before actually discussing the role of metonymy in the derivation of these implicatures, I would like to emphasize once more the fact that metonymy is by no means the only major factor involved in implicature. A fundamental factor which cooperates with metonymy and with general pragmatic principles (such as Rele. This mental journey may be entirely imaginary, if the reader has never actually traveled on that route, or it may consist of the invitation to recollect an earlier real journey undertaken by the reader on the same freeway between Denver and the Rockies (this mountain range is metonymically referred to as the Continental Divide). . This is the case with implicature c of sentence 1, which is implicit in implicature d of the same sentence (see below).

57

58

Antonio Barcelona

vance) to invite these implicatures is knowledge of the nature and the topic of the text (Graesser, Wiemer-Hastings and Wiemer-Hastings 2001). This knowledge creates a number of expectations that facilitate the operation of pragmatic principles and metonymy in the reader’s derivation of certain implicatures, even from a small stretch of language. These expectations very often include those concerning the author’s communicative intentions (Gibbs 1999: Ch. 7). Another obvious facilitating factor is the preceding co-text.

3.3.1 Sentence 1 (If you have ever driven west on Interstate 70 from Denver to the Continental Divide, you have seen Mount Bethel). This discourse-initial sentence is particularly important because it invites implicatures that reflect the main expectations regarding the kind of information to be provided by the rest of the text. These implicatures seem to be connected to each other, so that the first one seems to invite the second one and so forth. Therefore, while I will be able to include in this chapter only some of the implicatures derived from the rest of the text portion subjected to this study the discussion of the metonymic basis of all of these implicatures in the first sentence has been included in it. (a) “Mt Bethel is located somewhere close to Interstate 70.” Confirmation rate: 6:1. The writer states that if Traveler (T) has driven west from Denver to the Continental Divide on Interstate 70 (I–70), this is sufficient condition for concluding that T has seen Mt Bethel (MB), while driving along I–70. On the basis of this statement, the reader can draw the further inference (which is the inference the writer actually expects to suggest) that MB lies somewhere close to I–70. On what basis does the reader arrive at this inference? On the basis of his experiential knowledge of the seeing frame, which contains this “knowledge unit”: “For Viewer V (= T) to see Percept P (= MB) from the Viewer’s Vantage Point VP (= some point on I–70), P must be visible by V from VP. The basic condition for its visibility is that P should be relatively close to VP.”

An event for precondition metonymy, operating within the seeing frame on the basis of this knowledge unit, seems to trigger this part of the inference: v seeing p (event) for p being relatively close to vp (precondition). In other words, the fact that traveler/viewer T has seen percept MB activates the precondition that percept MB must be relatively close to T’s vantage point, I–70. That is, Mt Bethel must be relatively close to Interstate 70 for the reader (which is the imaginary traveler/ viewer) to be able to see it. . The insistence on the use of the modifier relatively in this case is due to the common knowledge that certain geographical features that are seen as being close may actually be very far from the viewer in terms of travelling distance; however, in these cases, they are necessarily close enough to be visible.


(b) “A description of Mt Bethel may be coming up.” Confirmation rate: 7:0. This inference is certainly facilitated by the reader’s awareness of the nature of the text initiated by this sentence. A natural expectation in this kind of texts is that the mountain appearing in the title of the hiking report will be the mountain where the hiking experience took place and that some description of its relevant features will be provided by the ensuing text. Therefore, the mention of Mt Bethel in the very first sentence does not simply implicate that this mountain has general, undifferentiated relevance for the rest of the text: the type of text already creates a strong bias as to what to expect from it. Apart from this general textual basis, implicature (b) is guided by a cause for effect metonymy whereby the conventional introduction to a description (initial event/cause) activates an expectation of the actual description (subsequent event/effect). A frequent, conventional way of announcing or introducing the actual description of an entity or event is to mention a perceptual experience of it. Examples are:

(1) I have seen Mt Bethel. It’s an interesting place. It has a perfect pyramid shape. Have you ever met Mary Smith? Oh, she’s the most attractive girl I’ve ever met. You would fall in love with her immediately. You have probably seen Mt Bethel. It’s a pyramid-shaped peak.

In these examples, the speaker first mentions to his/her addressee a real or imaginary perceptual experience of an entity, thus indirectly announcing the ensuing description of that entity, which will normally follow. This conventional introduction to the act of description thus causes and activates the expectation that the immediately following discourse units may include the actual description of the entity or event involved. In the text introduced by the sentence under analysis, the description of a number of identifying features of Mt Bethel is actually provided by sentences 2, 3 and 4. (c) “A series of travel or arrival instructions to Mt Bethel may be coming up.” Confirmation rate: 5:2. This implicature was only confirmed by five informants, but since it is implicit in implicature (d), which was accepted by six subjects, I have kept it as a valid inference. It is based on inferences (a) and (b) via metonymy and is also facilitated by the reader’s awareness of the type of text. By means of inference (a), the reader realizes that the first sentence provides a rough geographical location of Mt Bethel. By means of inference (b), the more or less conscious expectation is created in the reader that some description of Mt Bethel may be coming up. These two actions (namely, providing a rough geographical location of the destination and a general

59

60

Antonio Barcelona

description of it) are two important elements in any regular set of travel instructions. But these are not the only typical travel instructions indirectly provided by this sentence. The general travel route is also provided (westwards on Interstate 70 from Denver on the route to the Rockies) and the transport means is suggested (a motor vehicle, probably a car). All of these elements constitute the “preliminary travel information” section of what might be called the travel instructions frame. The ideal travel instructions frame contains at least the following basic types of instructions: a. preliminary travel information 1. Information on the general geographical location and a general description of the destination. 2. Instructions for choosing the correct general route. 3. Instructions for choosing the right means of transport. b. instructions for the central part of the journey 1. Instructions for initiating the journey. 2. Instructions for following the set route (which include taking into account important road signs, specific features of the area leading to the destination, etc.). c. instructions for the final part of the journey 1. Instructions for identifying the destination when approaching it and for accessing it. 2. Instructions for leaving the vehicle. The preliminary travel information section, as a salient part, is capable of metonymically evoking the whole travel instructions frame (part of frame for whole frame). This metonymy, thus, guides inference (c), again in cooperation with the reader’s awareness of the nature of the text. (d) “These travel instructions may be given in order to locate exactly the point at which the hiking experience reported in this text began.” Confirmation rate: 6:1. This overall inference may arise in the reader’s mind right after reading sentence 1, or it may arise later, perhaps after some or all of the remaining arrival instructions are given throughout the first paragraph of the text. Anyway, it is quite likely to arise right after reading sentence 1, given the reader’s knowledge of the nature of the text, which, as a guide for prospective hikers, already creates the strong expectation that the exact point where the hiking began is the specific destination of the mental journey on Interstate 70 and that this point will be located exactly for the reader (it is actually located explicitly by sentence 5 in conjunction with sentence 4). This expectation is activated and reinforced metonymically by the first sentence.


The use of an imaginary journey to Mt Bethel is an indirect strategy used to deliver a set of instructions for traveling to Mt Bethel. In the typical hiking report, the travel or arrival instructions are supposed to guide the reader to the initial point of the hiking route. By virtue of the metonymic connection between an action and its usual purpose, the mental activation of the action may lead to the mental activation of the purpose. Thus, the metonymy giving travel instructions (action) for guiding the traveler to the exact starting point of the hike (purpose) triggers, within what might be called the hiking report frame, the inference under discussion. The rest of the first paragraph, being longer, invites a larger number of text-level implicatures. Nearly all of these implicatures are directly guided or at least facilitated by metonymy. I will only have enough space to comment in some detail on the metonymic guidance/facilitation of a handful of the implicatures confirmed by my native speaker informants, and simply to indicate the metonymic basis of the rest. It must be borne in mind that, though partly guided or facilitated by metonymy, each of these implicatures also builds on those invited by previous sentences and is also facilitated by the awareness of the type of text.

3.3.2 Sentence 2 (Several miles (?) before approaching the exit for Hwy. 6 which leads to Loveland Pass you will see a pyramid shaped peak just to the north of the Interstate) (a–1) “The writer assumes that I (the reader) know and/or must be able to identify the location on Interstate 70 of the exit for Highway 6 which leads to Loveland Pass.” Confirmation rate: 7:0. This implicature is mainly based on a grammatical clue (the definiteness of the NP the exit [. . .] Loveland Pass). Therefore, it is apparently not guided by metonymy. (a–2) “The writer instructs me to travel mentally along Interstate 70 to the exit for Highway 6 and then to travel mentally back from this point for several miles.” Confirmation rate: 3:4. Rejected. (a–3) “The pyramid-shaped peak is probably very close to the endpoint of this backward path.” Confirmation rate: 3:4. Rejected. (a–4)

“The peak that has been described as a pyramid-shaped peak is probably Mt Bethel.” Confirmation rate: 7:0.

. However, the mention of Loveland Pass as the endpoint of a trajectory may be argued to constitute a metonymic reference point facilitating the activation (for those readers that have some knowledge of the area) of the initial point of the trajectory, namely the exact location of the entry to Highway 6.

61

62

Antonio Barcelona

This inference seems to be guided in part by the metonymic connection between a peak and a mountain. One of the major senses of the noun peak is “the pointed top of a mountain” (Oxford English Dictionary, henceforth abbreviated as OED, first part of sense 5 in the entry for this noun). A metonymic extension from this sense (also registered by the OED in the second part of sense 5) is “a mountain or hill having a more or less pointed summit, or of conical form.” Both senses are also registered in Webster’s. In sentence 2, it is not clear which of these two senses is being activated, but, in neutral contexts, basic senses (the “summit sense” in this case) are normally activated in preference to extended senses. If so, this activation is in turn the source for the metonymic activation of the mountain notion. That is, peak, as a part, metonymically activates mountain, as the whole. This metonymy is chained to another metonymy: category (mountain) for (contextually salient) member (mt bethel). The hypothetical identification between the pyramidshaped peak and Mt Bethel is thus guided in part by a metonymic chain: peak for mountain for mount bethel. This metonymic chain is prompted or at least facilitated by the text type (a mountain-hiking report) and by the previous sentence, where Mt Bethel has been mentioned. Further down in the text, in sentence 4, the noun peak is clearly used in the extended sense, i.e., as referring to a mountain with a pointed top. This sense is primed by the linguistic context, specifically the words As you pass the base of this peak on the highway. A driver cannot normally drive on a highway going by the base of the summit of a mountain, but he can quite frequently drive by the base (the foot) of a mountain. Therefore, the initial metonymy-based provisional inference that the pyramid shaped peak mentioned in sentence 2 is Mt Bethel is further reinforced in sentence 4. This reinforcement is also the result of the anaphorical connections between this peak (sentence 4), this roadway icon (sentence 3), and a pyramid shaped peak (sentence 2); some of these anaphorical connections are also facilitated in part by metonymy, as we see below in our comment on sentence 4. The valid inferences in the series (a–1)-(a–4) jointly lead to overall inference (a): (a) “The series of travel instructions continues, apparently with the purpose of locating exactly the point in Mt Bethel at which the hiking experience reported in this text began.” Confirmation rate: 6:1. This implicature builds on the two previous implicatures (a–1) and (a–4). It is guided in part by the metonymy part of frame for whole frame. The “Instructions for following the set route” section (represented here by the instructions for paying attention to the reference point consisting of the exit for Highway 6 (a–1)) and the “Instructions for identifying the destination” part (represented by the clues implicitly given to identify Mt Bethel (a–4)) are two sequential parts of the travel instruction frame (see above), both capable of evoking it.


Overall inference (a) leads (in my own reading) to overall inference (b): (b) “If so, the description of Mt Bethel begins now.” Confirmation rate: 3:4. Rejected.10

3.3.3 Sentence 3 (This roadway icon stands out like a sore thumb (cliché’) due to it’s prominence and near perfect triangular shape.) (a) “The roadway icon is the pyramid shaped peak just to the north of the Interstate.” Confirmation rate: 6:0.11 The sentence is connected to the preceding sentence, among other links, by means of the anaphorical noun phase This roadway icon, so that the reader automatically infers that this noun phrase refers to the same entity as the indefinite noun phrase a pyramid shaped peak just to the north of the Interstate in the previous sentence. The inference of this co-referential anaphorical connection is thus not fundamentally metonymy-guided. However, this roadway icon, without any further contextual support, might in principle also refer to other closely preceding NPs: the exit for Hwy. 6 which leads to Loveland Pass; Loveland Pass; and Hwy 6 – these might all in principle be metaphorically treated as “roadway icons.” The co-reference with a pyramid shaped peak is confirmed by the mention of two properties which metonymically activate the pyramid notion: salient property (triangular shape) for category (pyramid) and salient property (prominence) for category (icons). (b) “You can’t help noticing this peak.”12 Confirmation rate: 6:0. This implicature is invited by the sentence as a whole, and necessitates the previous identification of the peak just north of the Interstate as the referent of the NP this roadway icon. Therefore, it depends on the previous implicature invited by this sentence, namely (a). 10. One of the subjects entered a “???” symbol in this and in other YES slots for other inferences (see the relevant appendix), probably to indicate hesitation or surprise. These reactions are counted as disconfirmations of the implicature. 11. One of the subjects (distinct from the one entering ???) left blank the answer slots for the implicatures corresponding to sentence 3. He probably overlooked them, because he filled out all the slots for the implicatures derived from other sentences. This fact explains why the confirmation rates for the implicatures derived from this sentence affect six and not seven subjects. 12. According to one of my anonymous reviewers, a very similar implicature might also be invited by the first sentence, which, according to him, implicates that Mt Bethel is such a salient landmark that, if one travels west along Interstate 70 between Denver and the Rockies, one cannot help noticing it. Unfortunately, I had not realized this possible inference and did not test it with my informants. But if that reviewer were right, then that inference would now be reinforced by sentence 3 via implicature (b).

63

64

Antonio Barcelona

Implicature (b) is prompted by the attribution of the property prominent to this peak. The explicit attribution of this property is first carried out metaphorically by means of the fixed expression (which the author himself regards as a cliché) ‘stand out like a sore thumb’ and later literally by means of the expression thanks to it’s [sic]13 prominence and near perfect triangular shape. A prominent physical entity (irrespective of the dimension in which it is prominent) constitutes the “figure” in contrast with its physical environment, which constitutes the “ground” (Langacker 1987: 120–2). As such, it is very likely to be noticed (i.e. to be both perceived and consciously focused on) by an outside observer situated at a convenient vantage point (Langacker 1987: 123–6). There exists, then, a strong causal link between the prominence of an object and its very high likelihood (which in fact amounts to inevitability) of being noticed. When the assertion that entity X is prominent implicates that it is very likely, in fact inevitable, that it will be noticed, this implicature may be said to be facilitated by a metonymy: prominence of an entity (cause) for high likelihood for the entity to be noticed (effect). A consequence of this inference is the further provisional inference that the reader–imaginary traveler cannot help noticing Mt Bethel on the one hand due to the provisional identification of this mountain with the “pyramid shaped peak,” on the basis of sentence 2 (implicature a–4), and, on the other hand, the provisional identification of the “pyramid shaped peak” with the “roadway icon,” on the basis of sentence 3 (implicature a). (c) “The description of Mt Bethel apparently carries on.” Confirmation rate: 3:3. Rejected.

3.3.4 Sentence 4 (As you pass the base of this peak on the highway you will notice two very steep avalanche chutes which lead directly to the summit.) (a) “The words this roadway icon definitely refer to the same thing as the words a pyramid shaped peak.” Confirmation rate: 7:0. The anaphorical noun phrase this peak confirms the identification of the referent of the NP this roadway icon in sentence 3 with the referent of the NP a pyramid shaped peak occurring in sentence 2, which had in turn been provisionally identified with the referent of the NP Mt Bethel occurring in sentence 1. The immediate grammat13. The implicature that the writer is a less educated person is invited exclusively by the wrong spelling of the determiner ‘its’. This implicature does not seem to be relevant for the appreciation of text-internal cross-sentence connections nor does it seem to be invited by the text as a whole. Therefore it is not a text-level implicature and for this reason it has not been included in this study. The important role of metonymy in this and other sentence-internal implicatures in this passage is studied in Barcelona (2005) and in Barcelona (in preparation).


ical antecedent of this peak (metonymically a mountain) is this roadway icon. The anaphorical connection this peak – this roadway icon hinges not only on the use of the anaphorical definite determiner this, but also on the use of the term base, which metonymically evokes the notion of a triangle (part for whole). Since the “roadway icon” had been said to be triangular, talking now about the base of this peak-mountain implies that this peak-mountain is triangular and confirms that it is the same “thing” as the “roadway icon,” which had been said in sentence 3 to be triangular. Thus, the initial identification of the “roadway icon” with the “pyramid shaped peak” is now fully confirmed. (b) “The avalanche chutes are located on the mountain topped by this peak.” Confirmation rate: 6:1. This implicature has been treated as a text-level implicature because the avalanche chutes are mentioned again in sentence 6 and the identification of their location with respect to the peak mentioned earlier in the text affects the overall comprehension of the passage, not just of sentence 4. The inference is connected to the metonymic sense “mountain” of the lexeme peak and it is guided in part by two metonymies. Many mountains, especially very high ones, have avalanche chutes, which thus constitute a salient element that can metonymically activate the notion of a mountain: avalanche chute (salient part) for mountain (whole). A similar metonymy, prompted by the word summit, connects summit (salient part) to mountain (whole). Both metonymies activate the only mountain available in the immediate discourse space as the only likely location of the avalanche chutes: the mountain topped by the pyramid shaped peak, provisionally identified with Mt Bethel. Thus it can also be provisionally concluded that the avalanche chutes are located in Mt Bethel.

3.3.5 Sentence 5 (This is the route I chose for my ascent). (a) “The story of the writer’s climbing of the mountain may be about to begin.” Confirmation rate: 7:0. This inference is grounded on the statement that the writer chose the route consisting of the avalanche chutes anaphorically referred to by this. If a person tells someone else who has not asked specifically about it which route he chose to climb a mountain, he normally does so because he intends to tell his interlocutor the whole story of his ascent. The implicature is thus guided by the metonymy action (telling someone the route chosen for the ascent) for typical purpose (telling the story of the ascent). And, given the connection between the typical initial part of a story and the whole story, another metonymy probably also leading to the implicature is salient initial subevent (telling the route chosen for

65

66

Antonio Barcelona

the ascent) for complex event (telling the whole story of the ascent) (Kövecses and Radden 1998: 52). (b) “The writer intended to climb the peak immediately and he actually climbed it.” Confirmation rate: 5:2. Rejected. (c) “The peak is now definitely confirmed to be Mt Bethel.” Confirmation rate: 6:1. The initial inference drawn from sentence 2 that identified the “pyramid shaped peak just to the north of the Interstate” with Mt Bethel had received additional support from further inferences drawn from sentences 3 and 4. Sentence 5 definitively confirms this initial guess. Curiously enough, the name Mt Bethel is encountered, apart from the title, only once in the whole report, namely in sentence 1. If the text is, as stated in the web page frame, formally a report on a hike in Mt Bethel and if the author has been telling us that it is easy to get to Mt Bethel, and that at a certain point of our imaginary journey on Interstate 70 we will see a certain peak with certain features, etc... why should he bother to begin to tell us the story of his ascent to this peak if this peak were not actually Mt Bethel, i.e. the mountain that, according to the title of the hiking report, is the object of the hike? In other words, we apply the Principle of Relevance here to conclude that the mention of the writer’s past intention to climb the peak can only be relevant to the discourse topic if this peak is actually Mt Bethel. The inference is, then, prompted by this pragmatic principle. But the application of this principle is doubtless guided or at least facilitated by the metonymic connection between an event (an action in this case, namely, telling the story of a mountain hike) and its normal precondition, namely that the object of the hike can safely be identified as the mountain announced in the title of the report and in sentence 1: Mt Bethel. This metonymy may be called event for precondition. (d) “The earlier inference that these travel instructions are given in order to locate precisely the point on Mt Bethel at which the hiking experience reported in this text began is now confirmed or at least greatly reinforced.” Confirmation rate: 7:0. The confirmation by sentence 5 that the peak is Mt Bethel, and the information provided by this sentence that the hike followed a certain route on Mt Bethel (the two very steep avalanche chutes) jointly confirm the earlier implicature arrived at after sentence 1 that the travel instructions provided in the initial paragraph obey the purpose of (mentally) situating the reader at the exact point in Mt Bethel where the hiking experience began. This overall implicature is not directly guided by metonymy, but it is based on other implicatures which are guided or facilitated by metonymy. The exact point on Mt Bethel (i.e. the precise point on one of the avalanche chutes) from which the


hike began is specified in sentence 6 and in the second paragraph.

3.3.6 Sentence 6 (I pulled off the highway to begin my hike from where it looked as though there once may have been a parking area at the base of the westernmost avalanche chute and well off the highway right-of-way.) (a) “The writer got out of the vehicle in which he was traveling.” Confirmation rate: 5:2. Rejected. (b) “The story of the writer’s hike to the summit definitely begins now.” Confirmation rate: 7:0. The description of the place where the writer pulled off the highway and got off the car14 to begin his climb (“hike”) announces the story of the rest of the climb. This inference is guided by the metonymy sub-event (describing the place where he pulled off the highway and got out of the car in order to begin the hike) for subsequent complex event (telling the whole story of the climb). (c) “The purpose of this story is to provide a set of instructions for hiking in Mt Bethel.” Confirmation rate: 7:0. The inference occurs against the background of what might be called the mountain hiking instructions icm/frame. This cognitive model, like the travel instructions frame, has a number of basic parts (instructions for choosing at least an initial route for the hiking, instructions for getting to the initial point of that route, instructions for starting the hike, information on the obstacles and difficulties along the route and instructions to surmount them, information about the final destination of the hike and how to solve any possible problems there, and instructions for leaving the hiking area and getting to the initial point of the trip back home). Given the circumstances affecting the ascent reported in this text, which required the hiker to travel in a motor car to the hiking area, the “instructions for getting to the initial point of that route” include instructions on where to park the vehicle so as to be able to walk to the initial point of the hike. The mention of two of the sets of instructions included in the frame acts as a metonymic source for the whole frame. The instructions mentioned are those regarding where to park and leave the car and those for starting the hike (it should begin at the base of the western-most avalanche chute). The metonymy at work is part of a frame for the whole frame. 14. Surprisingly, this implicature (implicature (a) from this sentence) was confirmed by only five informants. Though implicit in implicature (b), I have not discussed its metonymic basis here for lack of space. It may described as initial sub-event (driving the car out of the highway) for whole event sequence (stopping or causing someone else to stop the vehicle and getting out of it)

67

68

Antonio Barcelona

From sentence 6 onwards, the rest of the story can be interpreted in part as delivering the rest of the set of instructions required by the mountain hiking frame.

3.4 Main metonymic types encountered in the study An interesting observation is the recurrence of certain generic types of conceptual metonymies as guides or facilitators of textual implicatures. In the present case study, these types are: event for precondition (imp (“implicature”) a, st (“sentence”) 1; imp c, st 5), cause for effect (imp b, st 1; imp b, st 3), part for whole (imp c, st 1; imp a–4, st 2; imp a, st 2; imp a, st 4; imp b, st 4; imp a, st 5; imp b, st 6; imp c, st 6), action for purpose (imp d, st 1; imp a, st 5), category for member (imp a–4, st 2), salient property for category (imp a; st 3). As can be seen, some of these generic types operate more than once even in the derivation of implicature in just one paragraph. In my earlier studies (Barcelona 2002b, 2003b, 2005), I have observed similar patterns of recurrence, not only in implicature derivation, but also in other types of meaning construction. The existence of all of these metonymies can be justified on independent grounds. They have been attested (under the same or a similar label, and either at a lexical-semantic level or at other levels) in my own published work on metonymy (e.g. Barcelona 2003b, 2005) and also in that of many other authors (e.g. Kövecses and Radden 1998; Fass 1997; Panther and Thornburg 2003).

4. Conclusions In this chapter, evidence has been presented for the fundamental role of metonymy in text-level implicature derivation. As stated in section 1, the research hypothesis of the case study reported in this chapter is that all or most of the text-level implicatures invited by the text portion chosen for the study are “guided” or “facilitated,” along with contextual factors and general pragmatic principles, by one or more conceptual metonymies; in other words, that metonymy is a fundamental factor in this type of meaning construction. Besides testing this hypothesis, another goal of the study was to investigate whether the ubiquitous role of metonymy in the derivation of textual implicatures that I had found in my previous studies of brief twosentence dialogues also occurred in a comparatively extended piece of narrative discourse like the text portion in this study. Both hypotheses have been fully confirmed by the present case study. Out of 16 textual implicatures confirmed by my informants, 14 can be claimed to be guided by metonymy, always in cooperation with general pragmatic principles, with the co-text, and with the reader’s awareness of the nature of the text. I had found in my


earlier case studies on dialogues (as reported in Barcelona 2002b, 2003b and 2005) a similar prevalence of metonymy in this implicature-guiding function. A distinctive mark of serious research is falsifiability. The results of this study can be falsified by showing that (i) the implicatures considered in it cannot in any case be derived, even considering the nature of the text chosen, from the text portion studied in it; and (ii) that the conceptual metonymies claimed to guide or facilitate the implicatures are proposed on an ad hoc basis (i.e. that they cannot be found in other areas) and that they are not consistent with my own definition of metonymy. However, the methodology followed ensured an unbiased selection both of the portion subjected to the study and of the implicatures under examination, and the metonymy types invoked in the analysis have been repeatedly invoked by other analysts and are consistent with my definition. Therefore, these results can be regarded as reliable and, in principle, representative for a portion (at least for the initial portion) of a relatively long narrative text. Therefore, this case study confirms the research hypothesis that, at least from a cognitive-linguistic perspective, metonymy is as powerful an implicature-guiding tool in relatively long written narrative texts as in short written dialogues. In Boroditsky’s (2000: 2) words, “strictly linguistic evidence can have only limited import for theories of mental representation as it would be scientifically imprudent to assume that patterns in language are necessarily a reflection of patterns in thought.” The investigation of the psychological reality of the results of this study and of others along the same lines would require extensive detailed psycholinguistic and cognitive-psychological experimentation, which is beyond my ability and means, but which would constitute a fascinating object of inquiry for experimental psycholinguists and cognitive psychologists.15

References Barcelona, Antonio. 2002a. Clarifying and applying the notions of metaphor and metonymy in cognitive linguistics: An update. In Metaphor and Metonymy in Comparison and Contrast, René Dirven and Ralf Pörings (eds), 207–77. Berlin: Mouton de Gruyter. Barcelona, Antonio. 2002b. On the ubiquity and multiple-level operation of metonymy. In Cognitive Linguistics Today [Łódż Studies in Language], Barbara Lewandowska-Tomaszczyk and Kamila Turewicz (eds) 207–24. Frankfurt: Peter Lang. 15. Gibbs (1994: 328–31) states that all the research on script-based language processing in psychology and artificial intelligence unknowingly illustrates the importance of metonymy in implicature. But no systematic psycholinguistic research of the role of metonymy in pragmatic inferencing has ever to my knowledge been carried out. An interesting piece of research in this direction is Gezgin (2006).

69

70

Antonio Barcelona

Barcelona, Antonio. 2003a. Metonymy in cognitive linguistics. An analysis and a few modest proposals. In Motivation in Language: Studies in Honor of Günter Radden, Hubert Cuyckens, Thomas Berg, René Dirven, and Klaus-Uwe Panther (eds), 223–55. Amsterdam: John Benjamins. Barcelona, Antonio. 2003b. The case for a metonymic basis of pragmatic inferencing: Evidence from jokes and funny anecdotes. In Panther and Thornburg (eds), 81–102. Barcelona, Antonio. 2005. The multilevel operation of metonymy in grammar and discourse with particular attention to metonymic chains. In Francisco J. Ruiz de Mendoza Ibáñez and Sandra Peña Cervel (eds), 313–52. Barcelona, Antonio. n.d. “The distinction between metaphor and metonymy: A question of (a)symmetry?”. Paper presented at the Fourth RAAM (Researching And Applying Metaphor) Conference. University of Manouba, Tunisia, 5–7 Apr. 2001. Barcelona, Antonio (in preparation). On the Pervasive Role of Metonymy in Constructional Meaning and Form and in Discourse Comprehension: A Corpus-Based Study from a Cognitive-Linguistic Perspective. (Provisional title.) Boroditsky, Lera. 2000. Metaphoric structuring: Understanding time through spatial metaphors. Cognition 17: 1–28. Brown, Gillian and Yule, George. 1983. Discourse Analysis. Cambridge: Cambridge University Press. Croft, Wiliam. 1993. The role of domains in the interpretation of metaphor and metonymy. Cognitive Linguistics 4(4): 335–70. Fass, Dan. 1997. Processing Metonymy and Metaphor [Contemporary Studies in Cognitive Science and Technology 1]. Greenwich, Connecticut and London: Ablex. Fauconnier, Gilles. 1997. Mappings in Thought and Language. Cambridge: Cambridge University Press. Fillmore, Charles. 1985. Frames and the Semantics of Understanding. Quaderni di Semantica 6(2): 222–54. Gezgin, Ulas Basar. 2006. Relationship of Bodily Communication with Cognitive and Personality Variables. PhD dissertation, Middle East Technical University, Laboratory for Computational Studies of Language. (Full text available at http://cogprints.org/4902/). Gibbs, Raymond W. Jr. 1994. The Poetics of Mind: Figurative Thought, Language, and Understanding. Cambridge: Cambridge University Press. Gibbs, Raymond W. Jr. 1999. Intentions in the Experience of Meaning. Cambridge: Cambridge University Press. Graesser, Arthur C., Wiemer-Hastings, Pieter and Wiemer-Hastings, Katja. 2001. Constructing inferences and relations during text comprehension. In Text Representation: Linguistic and Psycholinguistic Aspects, Ted Sanders, Joost Schiperoord and Wilbert Spooren (eds), 249– 71. Amsterdam: John Benjamins. Kövecses, Zoltán, and Radden, Günter. 1998. Metonymy: Developing a cognitive linguistic view. Cognitive Linguistics 9(1): 37–77. Lakoff, George. 1987. Women, Fire, and Dangerous Things: What Categories Reveal about the Mind. Chicago IL: Chicago University Press. Lakoff, George and Johnson, Mark. 1980. Metaphors We Live by. Chicago IL: University of Chicago Press. Lakoff, George and Johnson, Mark. 1999. Philosophy in the Flesh: The Embodied Mind and Its Challenge to Western Thought. New York NY: Basic Books.


Lakoff, George and Turner, Mark. 1989. More than Cool Reason: A Field Guide to Poetic Metaphor. Chicago IL: University of Chicago Press. Langacker, Ronald W. 1987. Foundations of Cognitive Grammar. Vol.1: Theoretical Prerequisites. Stanford CA: Stanford University Press. Langacker, Ronald W. 1993. Reference-point constructions. Cognitive Linguistics 4: 1–38. Oxford English Dictionary. 2002. CD–ROM edition of Second Edition (1989), and of Additions Series (1993–1997). T. J. Benbow, Main Editor. Oxford: Oxford University Press. Panther, Klaus-Uwe. 2005. The role of conceptual metonymy in meaning construction. In Ruiz de Mendoza Ibáñez and Peña Cervel (eds), 353–86. Panther, Klaus-Uwe and Thornburg, Linda. 1998. A cognitive approach to inferencing in conversation. Journal of Pragmatics 30: 755–69. Panther, Klaus-Uwe and Thornburg, Linda (eds). 2003. Metonymy and Pragmatic Inferencing [Pragmatics and Beyond New Series]. Amsterdam: John Benjamins. Papafragou, Anna. 1996. On Metonymy. Lingua 99(4): 169–95. Peirsman, Yves and Geeraerts, Dirk. 2006. Metonomy as a prototypical category. Cognitive Linguistics 17(3): 269–316. Radden, Günter. 2005. The ubiquity of metonymy. In Cognitive and Discourse Approaches to Metaphor and Metonymy, José Luis Otal Campo, Ignasi Navarro i Ferrando and Begoña Bellés Fortuño (eds), 11–28. Castellón: Jaume I University. Ruiz de Mendoza, Francisco José and Pérez Hernández, Lorena. 2003. Cognitive operations and pragmatic implication. In Panther and Thornburg (eds), 23–49. Sperber, Dan and Wilson, Deirdre. 1995. Relevance: Communication and Cognition (2nd ed.). Oxford: Basil Blackwell. Thornburg, Linda and Panther, Klaus-Uwe. 1997. Speech act metonymies. In Discourse and Perspective in Cognitive Linguistics [Current Issues in linguistic theory 151], Wolf-Andreas Liebert, Gisela Redeker and Linda Waugh (eds), 205–19. Amsterdam: John Benjamins.

Appendix 1. Complete text downloaded from http://hikingincolorado.org/beth.html: MOUNT BETHEL If you have ever driven west on Interstate 70 from Denver to the Continental Divide, you have seen Mount Bethel. Several miles (?) before approaching the exit for Hwy. 6 which leads to Loveland Pass you will see a pyramid shaped peak just to the north of the Interstate. This roadway icon stands out like a sore thumb (cliché’) due to it’s prominence and near perfect triangular shape. As you pass the base of this peak on the highway you will notice two very steep avalanche chutes which lead directly to the summit. This is the route I chose for my ascent. I pulled off the highway to begin my hike from where it looked as though there once may have been a parking area at the base of the western-most avalanche chute and well off the highway right-of-way. As I was following no trail I began by heading straight up this (western) chute (on the north face). Footing was good and I had no problem bushwhacking my way towards tree line. About two-thirds of the way to the tree line I found a game trail which proved to be helpful, yet steep. I followed this game trail up through a rocky area just below tree line. Once through this rocky area the slope temporarily leveled out looking somewhat like an alluvial fan.

71

72

Antonio Barcelona

This is about where the previous week’s snow began to cling to the shady areas and in the deeper ravines. I still had no problem with my footing and continued to climb towards the summit. About 400 feet from the summit I was forced to head to the east so that I could bypass a short but very steep technical climb. I found hiking this ridge to the summit to be relatively effortless (no snow, good footing, stable rocks, some grass) although steep. I reached what I thought to be the summit overlooking the highway only to find the actual summit was 4 peaks and several hundred yards of flat walking to the north - away from the highway. This is where I found a special guest book reserved exclusively for obscure summits! The view from the top included (in my neighborhood) Paiute Peak, Hag Mountain, The Loveland Loop, Loveland Pass and Torreys-Peak. Also visible were Mount Evans (to the southeast) and Berthoud Pass to the north-northeast. As it was late September with new snow around me, ominous clouds about a thousand feet over my head, and chilly temperatures magnified by intermittent gusts of wind, I decided to begin my journey back to the car and the highway below. Note Regarding Sounds and the Highway: For the better part of my journey the sound (loud near the bottom and muffled near the top) of the Interstate was a constant reminder of civilization. Once over the top and away from the highway these sounds disappeared. I chose a slightly different route down (for safety reasons). This included descending further down the northeast ridge to a point lower than where I originally accessed this line. This took me to the easternmost avalanche chute on north face of the mountain. Well below tree line I found myself in the remains of what must have been an avalanche from last season. There was loose dirt, whole trees ripped out of the ground and pieces of trees that must have broken like match sticks (I know - another cliché´). From this point it was an easy trek back to my car. During this entire hike I encountered no other people and limited wildlife. I saw only Pica, Marmot and lots of Mule Deer scat. Overall this was a steep and moderately difficult hike but it was also very accessible as I was only away from Denver for around 6 hours! At this time I cannot give you an accurate measurement of the elevation gain but I’m guessing that it was around 2,500 feet. Along on this hike with me: No one.

2. Informal experiment with native speakers The experiment consisted of administering the questionnaire below to seven native speakers of English (four of them British, one of them Irish, and two American). As can be seen, the questionnaire already directs their attention to the implicatures. Therefore the experiment is not supposed to find out the implicatures they would have drawn if only the contextual information had been suggested to them. The purpose of the questionnaire is to find out which ones of my own set of implicatures they would find consistent with the text.

Contextual information given to subjects You are going to answer a few brief questions on the various sentences of the first paragraph in a text. The text is a report of a trip to a Colorado mountain with the purpose of hiking in it. The text forms part of a series of reports of similar trips to other Colorado mountains, available at (http://hikingincolorado.org/beth.html) (PLEASE DON’T OPEN THAT PAGE YET). This series is intended to be a source of information for prospective hikers in the area. So the reader can


naturally expect this text to be about Mt Bethel (which figures prominently in the title of the report), to situate this mountain and at least the beginning of the hiking route geographically, to provide a brief description of the shape and general appearance of the mountain and, of course, to provide a story of the arrival and the subsequent hiking experience of the narrator.

Instructions Please read the following sentence (the first sentence in the paragraph under examination). Don’t look at the rest of the paragraph yet: [1] If you have ever driven west on Interstate 70 from Denver to the Continental Divide, you have seen Mount Bethel. Now imagine you are a frequent reader of this sort of texts (as explained above, hiking reports), and confirm (put an X under “YES”) or disconfirm (put an X under “NO”) whether you would draw the following inferences (that is, the “ideas” that “come to your mind”) right after this sentence, taking into account the nature of the text and the sentence itself: INFERENCES

YES NO

a. “Mt Bethel is located somewhere close to Interstate 70.” b. “A description of Mt Bethel may be coming up.” c. “A series of travel or arrival instructions to Mt Bethel may be coming up.” d. “These travel instructions may be given in order to locate exactly the point at which the hiking experience reported in this text began.” Now read the whole paragraph under analysis (which is part of the larger text that you can read after the “experiment” at the website mentioned above). The spelling errors are in the original paragraph (I have numbered the sentences for ease of reference): [1] If you have ever driven west on Interstate 70 from Denver to the Continental Divide, you have seen Mount Bethel. [2] Several miles (?) before approaching the exit for Hwy. 6 which leads to Loveland Pass you will see a pyramid shaped peak just to the north of the Interstate. [3] This roadway icon stands out like a sore thumb (cliché’) due to it’s prominence and near perfect triangular shape. [4] As you pass the base of this peak on the highway you will notice two very steep avalanche chutes which lead directly to the summit. [5] This is the route I chose for my ascent. [6] I pulled off the highway to begin my hike from where it looked as though there once may have been a parking area at the base of the western-most avalanche chute and well off the highway right-of –way. Take a look again at sentence 2, and confirm or disconfirm whether you would draw (among others) these inferences from this sentence, taking into account the previous sentence and the nature of the text:

73

74

Antonio Barcelona

INFERENCES

YES NO

a-1. “The writer assumes that I (the reader) know and / or must be able to identify the location on Interstate 70 of the exit for Highway 6 which leads to Loveland Pass.” a-2. “The writer instructs me to travel mentally along Interstate 70 to the exit for Highway 6 and then to travel mentally back from this point for several miles”. a-3. “The pyramid-shaped peak is probably very close to the endpoint of this backward path”. a-4. “The peak that has been described as a pyramid-shaped peak is probably Mt Bethel a. “The series of travel instructions continues, apparently with the purpose of locating exactly the point in Mt Bethel at which the hiking experience reported in this text began.” b. “If so, the description of Mt Bethel begins now.” Take a look again at sentence 3, and confirm or disconfirm whether you would draw (among others) these inferences from this sentence, taking into account the previous sentences and the nature of the text: INFERENCES

YES NO

a. “The roadway icon is the pyramid shaped peak just to the north of the Interstate”. b. “You can’t help noticing this peak”. c. “The description of Mt Bethel apparently carries on.” Take a look again at sentence 4, and confirm or disconfirm whether you would draw (among others) these inferences from this sentence, taking into account the previous sentences and the nature of the text: INFERENCES

YES NO

a. “The words this roadway icon definitely refer to the same thing as the words a pyramid shaped peak” b. “The avalanche chutes are located on the mountain topped by this peak.” Take a look again at sentence 5, and confirm or disconfirm whether you would draw (among others) these inferences from this sentence, taking into account the previous sentences and the nature of the text: INFERENCES a. “The story of the writer’s climbing of the mountain may be about to begin”. b. “The writer intended to climb the peak immediately and he actually climbed it”. c. “The peak is now confirmed to be Mt Bethel.”

YES NO


d. “The earlier inference that these travel instructions are given in order to locate precisely the point on Mt Bethel at which the hiking experience reported in this text began is now confirmed or at least greatly reinforced.” Take a look again at sentence 6, and confirm or disconfirm whether you would draw (among others) these inferences from this sentence, taking into account the previous sentences and the nature of the text: INFERENCES a. “The writer got off the vehicle in which he was traveling.” b. “The story of the writer’s hike to the summit definitely begins now.” c. “The purpose of this story is now confirmed to be that of providing a set of instructions for hiking in Mt Bethel.”

YES NO

75

chapter 4

Chained metonymies in lexicon and grammar A cross-linguistic perspective on body part terms* Martin Hilpert Rice University

This chapter investigates chained metonymies, which are metonymies that involve multiple conceptual shifts. Drawing on a survey of body part terms in a large balanced sample of languages, it explores which types of conceptual shifts give rise to extended lexical meanings, and which types are involved in the development of more schematic, grammatical meaning. It is shown that for body part terms, chained metonymies mostly lead to lexical extensions, while serial extensions that are rooted in metaphor are cross-linguistically common in the development of grammatical meaning. Keywords: body part terms, chained metonymies, cross-linguistic comparisons, grammaticalization, interaction of metaphor and metonymy, lexicalization, metaphor, metonymy

1. Introduction This chapter investigates the construction of meaning through chained metonymies, which are metonymies that involve multiple conceptual shifts. Interest in the serial nature of metonymy goes back at least to Reddy (1979), who observed that expressions such as example (1) involve several metonymic mappings.

(1) You’ll find better ideas than that in the library. (Reddy 1979: 309) ideas → words → pages → books

Reddy argues that hearers understand sentence (1) by inferring that ideas are expressed in words, printed on pages within books, which are found in libraries. More recently, several studies have taken up Reddy’s observation (Nerlich and Clarke 2001; Geeraerts 2002; Ruiz de Mendoza and Diéz 2002; Barcelona 2003), show * I wish to thank Thomas Berg and Günter Radden for their many helpful comments and their patience. Thanks are also due to the DAAD. Last but certainly not least, I would like to thank Klaus-Uwe Panther, who fostered my interest in the topic of this chapter. . Linguistic examples in this chapter are cited along with their published source. If such a reference is absent, the example has been constructed by the author.

78

Martin Hilpert

ing its continuing relevance for cognitive linguistics. While these studies mostly focus on lexical chained metonymies as found in idioms and other figurative expressions, a different strand of cognitively oriented work investigates the role of chained metonymies in grammar (Heine et al.1991; Bybee et al. 1994; Traugott and Dasher 2002). For example, Heine and Kuteva (2002: 129) suggest that Bambara nyέ ‘eye’ has grammaticalized into the temporal deictic marker ‘before’ through a chain of semantic shifts, as shown in (2).

né nyέ (2) à nà-na 3sg come-past 1sg before ‘She arrived before me.’ (Heine and Kuteva 2002: 129) eye → face → front → before

This chapter aims to integrate the two approaches and analyzes chained metonymies as a phenomenon that pertains to both lexicalization and grammaticalization. Such an approach needs to ask whether there are systematic differences between semantic extensions that lead to lexical meaning and those extensions that lead to grammatical meaning. It is argued that such differences do in fact exist. The present analysis addresses the above question by using data from dictionaries to explore semantic extensions of body part terms across languages. Mappings that give rise to lexical and grammatical meaning respectively are collected from different languages, which are then compared. Body part terms are investigated in this chapter because they have been identified as a productive source of extensions, both lexical (Niemeier 2000; Deignan and Potter 2004; Hilpert 2006a, inter alia) and grammatical (Heine et al. 1991; Hollenbach 1995; Matsumoto 1999; Svorou 1994, inter alia). For example, in languages such as Khwe (Kilian-Hatz 2003), the equivalent of the English word back has secondary senses such as ‘behind’ and ‘after’ which illustrate well-known grammaticalization processes. Lexical semantic extensions are exemplified by the Kanuri equivalent of English arm, which has the secondary sense ‘elephant trunk’ (Cyffer and Hutchison 1990). In Wardaman, belly can denote ‘kangaroo pouch’ (Merlan 1994). Most meaning extensions of body part terms can be shown to have a clear motivation through either metaphor or metonymy, as has been argued in many studies before (Allan 1995; Bowden 1991; MacLaury 1989, inter alia), which makes body lexis a fruitful point of departure for the study of chained metonymies. Through the case study of body part terms, this chapter also addresses the question whether, given a sequence of conceptual mappings, different mappings tend to occur in different positions. Previous analyses (Goossens 2002; Taylor 2002) . The last step in the semantic development is actually not metonymic, as it involves the conceptual metaphor time is space (Lakoff and Johnson 1980).

Chained metonymies in lexicon and grammar

have argued on the basis of English data that metonymic mappings tend to precede metaphoric mappings, but as yet these claims have not been empirically tested against cross-linguistic data. This chapter is thus also intended as a contribution to the ongoing discussion about the interplay of metonymy and metaphor in the construction of meaning. The structure of the chapter is as follows. Section 2 introduces the notion of chained metonymies. Section 3.1 outlines the cross-linguistic methodology adopted here, 3.2 reports on the collected serial extensions that map body part terms onto lexical meanings, 3.3 presents the extensions that involve grammatical meanings, and 3.4 organizes these findings into generalizations regarding common sequences of metaphors and metonymies. Section four concludes that the claims of Goossens (2002) and Taylor (2002) hold true for lexical extensions, but need qualification for grammatical extensions.

2. Chained metonymies To introduce the notion of chained metonymies, a brief definition of metonymy is in order. In cognitive linguistics, metonymy is viewed as a conceptual phenomenon, rather than a mere substitution of one word for another. Radden and Kövecses (1999: 21) thus define metonymy as the conceptual link between two entities in the same frame of reference: “Metonymy is a cognitive process in which one conceptual entity, the vehicle, provides mental access to another conceptual entity, the target, within the same idealized cognitive model.” This chapter uses the taxonomy of metonymic relationships proposed by Radden and Kövecses to analyze meaning extensions. The aim of the analysis is to find regularities in cross-linguistically common sequences of conceptual steps that lead to lexical and grammatical meaning. Much like metaphor, metonymy is ubiquitous in language, and in fact in general reasoning (Gibbs 1999; Panther 2005). Common metonymic mappings such as part for whole or cause for effect underlie everyday expressions like (3) and (4), which mean more than they literally state.

(3) We need some new faces around here. (Lakoff and Johnson 1980) body part → person (4) General Motors had to stop production. (Panther and Thornburg 2003: 11) obligation to act → action

In (3), a body part stands for an entire person. Example (4) means that production was actually stopped, hence the obligation to carry out an action stands for the action itself. These conceptual shifts are achieved by metonymic mappings, which

79

80

Martin Hilpert

Panther and Thornburg view as “reasoning principles in utterance interpretation” (2004: 94). Metonymic mappings can thus be understood as conceptual routines that guide the construction of meaning, on a par with pragmatic inference-guiding principles (e.g. Levinson 2000). Chained metonymies involve more than one conceptual shift. In example (5), one single metonymic mapping seems insufficient to account for the meaning that is constructed from the utterance. A rough paraphrase of (5) would be that Bob presented some interesting ideas.

(5) Bob gave an interesting paper. material → writing → ideas

Nothing in principle disallows a single metonymic mapping that has the material ‘paper’ directly stand for ‘ideas’. However, the chained metonymy suggested in (5), which maps ‘paper’ onto ‘writing’, which in turn stands for the expressed ‘ideas’, has several theoretical and empirical advantages (Hilpert 2006b). A theoretical argument in favor of chained metonymies is that these chains break up complex conceptual mappings into simple, well-motivated mappings with a strong experiential basis. The associations of ‘paper’ with ‘writing’ and ‘writing’ with ‘ideas’ emerge from everyday experience. Conversely, the association of ‘paper’ and ‘ideas’ is indirect, mediated only through the experience of reading and writing. A similar point is made by Grady (1997: 287), who advocates the decomposition of complex – and hence fairly abstract – metaphors into basic metaphors with a stronger experiential motivation. Whenever a metaphor is posited, Grady argues, it needs to be experientially motivated. This constrains the types of conceptual relationships that can be posited, and thereby enhances their explanatory power. While these theoretical considerations motivate an analysis in terms of chained metonymies, they do not satisfactorily answer the question of how many links there should be in a chained metonymy. To justify intermediate links between vehicle and target, empirical evidence needs to be taken into consideration. The first empirical constraint is that all intermediate links should represent attested, authentic expressions. The chained metonymy in (5) presupposes the existence of examples such as (6), in which paper stands for ‘writing’.

(6) Bob found a typo in his paper. material → writing

The constraint of productive intermediate links thus requires independent evidence for each metonymic mapping that is posited. If such evidence is sparse or absent, a single metonymic mapping should be preferred as the more parsimonious analysis. A second empirical constraint concerns the polysemy that is found with certain metonymic expressions. The chained model predicts that polysemous expressions


convey meanings that correspond to adjacent links in the chained metonymy. Consider the following examples.

(7) ‘Where do you come from?’ asked Betty with an eye on Lydia. (BNC G0X) eye → vision

(8) Chief executive Paul Hanson, with an eye on falling consumer demand in the UK, hopes to double exports by the end of the financial year. (BNC A7T) eye → vision → attention

(9) Two American companies with an eye on available and inexpensive manpower have set up software export units in the city of Bangalore, India’s own Silicon Valley. (BNC A8R) eye → vision → attention → desire

The English expression with an eye on NP is polysemous, conveying ‘vision’ in (7), ‘attention’ in (8), and ‘desire’ in (9). The chained model naturally accounts for this polysemy, since people are watching the things they pay attention to, and pay attention to the things they desire. An analysis in terms of three individual metonymic mappings would fail to capture this generalization. Again, experiential motivation favors the chained model. A third empirical constraint on chained metonymies is particularly relevant to the methodology of this chapter. The chained metonymy in (9) predicts a cross-linguistic implicational hierarchy. Specifically, it makes the prediction that if a language has a meaning extension of a body part term such as eye to the concept ‘desire’, the body part term should also have been extended to the meanings of ‘vision’ and ‘attention’. We should thus find expressions in that language in which eye is used to convey ‘vision’ and ‘attention’, respectively. If such expressions cannot be found, or sound unacceptable to native speakers, the analysis in terms of a chained metonymy is doubtful. To summarize, while there is broad agreement that chained metonymies do in fact exist, the respective analyses need to be constrained. Experiential motivation, productivity of intermediate links, polysemy across adjacent links, and cross-linguistic attestation of intermediate links are the constraints that guide the present analysis.

3. Body part terms and their chained metonymies 3.1 Methodology This study explores semantic extensions of body part terms with the aim of identifying cross-linguistically common serial metonymic mappings. Special interest is

81

82

Martin Hilpert

devoted to the distinction between lexical and grammatical meaning, and how different mappings may give rise to each respective type. It goes without saying that the distinction between lexical and grammatical is not categorical, but is inherently gradient (Hopper 1987). The continuum from lexical to grammatical meaning can be understood as a cline of increasing schematicity (Langacker 1987). For the present purposes, grammatical meaning is defined as the instantiation of a functional category such as aspect or modality, as well as deictic reference to temporal, spatial, and interpersonal relations (Heine and Kuteva 2002: 15). By exclusion, lexical meaning is defined as the remaining semantic space. The starting point of the study is the list of sixteen body part terms given in (10). (10) arm, back, belly, buttocks, ear, eye, face, finger, foot, forehead, hand, head, heart, jaw, mouth, tongue The equivalents of these English terms were looked up in bilingual dictionaries of 76 languages which represent a stratified probability sample of the world’s languages. The sample represents all known language families and subbranches in proportion to their respective sizes. This chapter adopts the general sampling scheme of Bybee et al. (1994: 311), but exchanges languages from the same language family subbranches (such as Niger-Kordofian or Aztec-Tanoan), wherever this leads to a more comprehensive dictionary resource. These replacements retain the weighed representation of the language families and therefore do not distort the balance of the sample. For each lexical entry, it was determined whether the body part term was used to convey secondary senses, which were then pooled in a database. To illustrate, a dictionary entry from Ma’di (Blackings 2000: 68) is shown in (11).

. As a typographical convention, body part terms such as back or eye are rendered in italics when they refer to a linguistic form of either English or some other language. Semantic extensions of body part terms, such as ‘behind’ or ‘attention’ are rendered in ‘single quotes.’ . The languages sampled are: Aitchin, Alabama, Anejom, Anywa, Awa, Balti, Bantawa, Basque, Bokobaru, Busa, Carolinian, Cayuga, Chantyal, Chechen, Danish, Delaware, Efik, Finnish, Ge’ez, Greek, Guaraní, Hani, Hausa, Hiri Motu, Hopi, Ilocano, Inuktitut, Iraqw, Kanuri, Karok, Kayardild, Khwe, Koiari, Kolami, Kongo, Koyukon Athabaskan, Kristang, Krongo, Kurdish, Kwoma, Kyaka Enga, Lushai, Ma’di, Maidu, Mandarin, Mara, Marshallese, Nandi, Nez Perce, Ngizim, Oneida, Pahlavi, Piro, Puget Salish, Rendille, Rhade, Sedang, Selepet, Senoi, Shona, Southern Sierra Miwok, Spanish, Tagalog, Tahitian, Thao, Tohono O’odham, Tümpisa Shoshone, Turkish, Uzbek, Vietnamese, Wardaman, Yir-Yoront, Yogad, Yoruba, Yugambeh, and Zapotec. Full bibliographical reference to the dictionaries and a key to the genetic affiliations of the above languages can be found at http://www.ruf.rice.edu/~hilpert /metonymy.doc. The reference section of this chapter only includes sources that are cited in the body of the text.


(11) mī noun 1. eye, 2. a haunting spirit: mī nā ká nyì ko rá the spirits will haunt or afflict you, mī àko blind, mī āko blindness, mī ‘bí eyelash, mī īnggwε flirtation, mī īnì seriousness. From this entry, the semantic extensions of eye for Ma’di are entered in the database as ‘spirit,’ ‘blind,’ ‘flirtation,’ and ‘seriousness.’ The meaning ‘eyelash’ is disregarded because its component parts are semantically transparent. By contrast, the component parts of mī àko ‘blind’ literally mean ‘without eyes,’ such that the targeted meaning is more specific than what is literally stated. Likewise, the component parts of mī īnggwε ‘flirtation’ literally mean ‘bright eyes,’ which also underspecifies the targeted meaning. While these secondary senses are metonymically motivated, none of them appear to be related to each other through a chained metonymy. Also, none of them convey grammatical or functional meaning. The Ma’di term for back (Blackings 2000: 81), however, exhibits grammatical meaning extensions that suggests a chain of semantic extensions. The lexical entry is given in (12). (12) ògū ‘bārá nikā ògū gá ògū nā opí ‘i ògū ògū

noun the back of a thing; back. carrying a child on the back, Opi was born after her. postposition behind, at the back. adverb next to come; after this or that.

From this entry, the semantic extensions ‘back part,’ ‘behind,’ and ‘after’ are entered into the database. Arguably, these senses are closely related. It has been argued repeatedly that body part terms such as back first develop into object part lexemes, then grammaticalize into deictic spatial markers, and from there on acquire temporal meaning (Heine et al.1991: 66, Matsumoto 1999: 22, Svorou 1994: 90). The proposed sequence of meaning extensions is shown in (13). (13) ògū nā opí ‘i (Blackings 2000: 81) ‘Opi was born after her.’ back → back part → behind → after These semantic changes are motivated by metaphor and metonymy. The first step involves the metaphor objects are humans, such that inanimate objects can be said to have a back. This conceptual metaphor is traditionally known as personi. Svorou (1994) also discusses developments in which environmental landmarks such as sky or earth and animal body parts (as opposed to human body parts) develop into deictic spatial categories. For an alternative account see Wilkins (1993: 11), who presents semantic developments from objects to body parts (such as bag → belly in Bantu languages), i.e. in the reverse direction.

83

84

Martin Hilpert

fication. The next step is metonymic, being based on the contiguity relationship between a part of an object and an adjacent region towards which it is oriented. This mapping can be called the part for region metonymy. The final step in the chain is again metaphoric, drawing on the metaphor time is space. The chain of meanings in (13) suggests that if a language has a meaning extension of the body part term back to the temporal concept ‘after,’ the meanings of ‘behind’ and ‘back part’ should also be present. The lexical entry from Ma’di dovetails with the sense development as proposed by Heine and colleagues, but it is an empirical question whether all languages with a temporal meaning of back display the same polysemy, or whether some languages actually derive the meanings ‘behind’ or ‘after’ directly from the human body part. While data from dictionaries provide much useful information, they cannot replace knowledge of a language’s history. It is hence beyond the scope of this chapter to discuss grammaticalization processes that were accompanied by morphophonological reduction. Cases in which the morphological substance of a body part term has been altered, or in which it has been reduced to a grammatical affix go unnoticed by the current methodology unless the dictionary compilers included a reference under the entry of the full lexical form. Another caveat concerns the fact that dictionary compilers may have left out a secondary sense which actually exists in the language. Since this is a realistic possibility, generalizations in this chapter will not be made from the singular presence or absence of entries in individual dictionaries, but preferably from convergent evidence that reflects the characteristics of more than one language. As a last qualification, dictionaries do not offer much information about the syntactic behavior of their entries. Corpus-based studies of figurative language (Deignan and Potter 2004; Hilpert 2006a) have shown that collocation and colligation patterns are instrumental in the disambiguation of lexical items. Information of this kind is not provided here. Despite these caveats, a comprehensive collection of meaning extensions of body part terms yields potentially instructive insights in three ways. First, it makes it possible to determine what lexical and grammatical concepts are typically targeted, and which of these targets tend to be co-present in the languages investigated. Second, the observed implicational hierarchies show that some metonymic extensions are semantically dependent on other extensions, thereby suggesting a chained me. Heine et al. (1991: 123) characterize the mapping of body part terms onto locative meaning as metaphorical, and invoke an ‘object to space metaphor’ as the underlying motivation. This study takes the viewpoint of Svorou (1994: 90), who advances an analysis in terms of metonymy. The relationship of an object part and its adjacent regions is based on contiguity rather than similarity, which motivates an account in terms of metonymy. Svorou proposes a chain of meaning extensions from ‘human body part’ to ‘relational part of object’ to ‘location adjacent to object part’ to ‘location in region of object part’.


tonymy. Finally, the pool of different chained metonymies encountered in the database can be used to draw generalizations about the nature of serial metonymy and meaning extension in general.

3.2 Extensions onto lexical meanings Virtually every language investigated in the survey exhibits a rich set of semantic extensions of body part terms, underscoring the importance of the human body for lexical structure (Lakoff and Johnson 1999). Most extensions are readily motivated through metaphor and metonymy. Very frequently, the term for eye refers to ‘vision’ through the instrument for action metonymy, and arm refers to ‘branch’ through the plants are humans metaphor. With some body parts, it cannot be easily decided whether a secondary sense is truly an extension, or whether the term is simply vague to begin with. In twenty-six languages of the sample, the term for hand can refer to ‘arm’ and vice versa. In another twenty, the term for finger also means ‘toe.’ In forty languages, the term for foot can refer to a ‘leg.’ While this kind of polysemy can be explained in terms of metonymy, it is hard to determine which sense is more basic and which is the extension. For the present purposes, these body part terms are thus assumed to be vague. Table 1 presents the most frequent semantic extensions of body part terms onto lexical concepts. Table 1. The most frequent lexical extensions from body part terms body part

extension (in number of languages)

arm back belly buttocks ear eye face finger foot forehead hand head heart jaw mouth tongue

hand (31), sleeve (12), branch (11), wing (9), handle (7), shoulder (5), strength (5) back part (42), spine (8), to turn (7), support (7), last (5) pregnancy (15), inside part (9), defecate (4), diarrhea (4), character (4) anus (15), back part (7), bottom part (4), hip (4) hearing (30), deaf (24), handle (6), attention (6), ignore (6), mushroom (5) vision (39), blind (17), attention (14), glasses (7), tear (7), jealous (6), knot (5) front (15), countenance (7), appearance (6), expression (5) toe (20), hand (10), to point (5), ring (4), arm (3), measure (3) leg (40), measure (11), footprints (10), walk (9), base (8), wheel (6), step (6) front (10), brow (6), face (5), top (3), cliff (3) arm (26), handle (7), finger (7), power (6), sleeve (6), help (5) top part (19), hair (11), intellect (10), beginning (10), chief (10), summit (8) emotions (26), character (8), core part (7), center (6), courage (6), mind (6) chin (17), cheek (8), slope (2) speech (32), opening (19), edge (11), speech act (9), entrance (8), beak (7) speech (26), lick (7), speech act (6), blade (2), word (2)

. This conceptual metaphor is one of the few which are reversible. An English example of human beings are plants would be I have my roots in the Chicago area.

85

86

Martin Hilpert

Not all extensions in Table 1 involve a series of conceptual mappings. The following sections narrow the focus to four conceptual mappings that are productive first steps for serial meaning extensions. First, body parts are frequently associated with their perceptual function. As a further extension of that, different aspects of perception may be targeted. A second common extension views body parts as instruments that are used for a particular action. Again, different aspects of this action are highlighted in subsequent extensions. A third strategy is the conceptualization of body parts as containers, which also allows further extension. Lastly, body parts often serve as a model to conceptualize the parts of physical objects. This, in turn, often leads to further extensions.

3.2.1 The organ of perception for perception metonymy The organ of perception for perception metonymy is frequently found with the terms for ear and eye. After being extended onto ‘hearing’ and ‘vision’ respectively, both are mapped onto the target ‘attention,’ as well as a range of other concepts. Thirty languages of the sample map ear onto ‘hearing.’ Many of these further extend the meaning onto more specific lexical concepts that stand in contiguity relationships with auditory perception. In eighteen languages, the body part term ear stands for the concept of ‘attention.’ This is arguably more specific than ‘hearing,’ since it involves deliberate action on the part of the perceiver. Sweetser (1990) has shown that perception verbs regularly develop into verbs of cognition, as well as into the lexical concepts of obedience, likes, and dislikes. The concept of ‘paying attention’ needs to be viewed as a semantic extension, since it need not actually involve auditory perception. In English, one can lend an ear to the needs of the community without necessarily being able to hear. The perception for attention metonymy thus maps an involuntary process onto an action which is deliberate and controlled. The mirror image of this meaning extension is the target ‘disregard,’ which is precisely the absence of attention. It is found in Hausa, Kristang, Lushai and Tagalog. The same metonymies apply, but the interpretation is inversed. An alternative possibility to derive this meaning would be from the common lexical extension ‘deaf,’ which occurs in twenty-four languages. However, of the four languages mentioned only Lushai has this sense of ear, which makes this derivation an unlikely possibility. The languages Balti, Hani, Kyaka Enga, and Selepet map ear onto ‘obedience,’ which is motivated through the cause for effect metonymy. The result of ‘obedi. The languages that extend ear to ‘attention’ are Balti, Busa, Chechen, Danish, Finnish, Greek, Hani, Hausa, Ilocano, Iraqw, Khwe, Kristang, Kurdish, Piro, Rendille, Tagalog, Turkish, and Yogad.


ence’ shows that the perception of a command has a tangible effect (Sweetser 1990). In five languages, ‘hearing’ is further extended to mean ‘hearsay.’ Here, an action stands for an associated object, which in this case is that which is perceived . The action for object metonymy (Radden and Kövecses 1999: 37) underlies for example English de-verbal nouns such as a drink or a bite, and can be viewed as the motivation for this particular extension. A schema of the chained metonymies discussed above is shown in (14). (14) ear → hearing → {attention, disregard, obedience, hearsay} Thirty-nine languages of the sample associate eye with ‘vision’ through the organ of perception for perception metonymy. Two lexical extensions appear to be conceptually dependent on this mapping, extending it to ‘attention’ and ‘beauty’ respectively. In fourteen languages, the term for eye stands for ‘attention.’10 As in the example of auditory perception, the concept of ‘vision’ maps onto ‘attention’ through the perception for attention metonymy. There is considerable overlap between the languages that target this meaning via either of the two perceptual organs. Eleven languages use both ear and eye to denote ‘attention.’ In Basque, Bokobaru, and Busa, expressions with eye convey the meaning ‘beauty.’ The English expression eye candy may serve as a comparison here. The notion of a perceived quality presupposes the idea of perception, and hence the perception for thing perceived metonymy (Radden and Kövecses 1999: 38). These chained metonymies are schematized in (15). (15) eye → vision → {attention, beauty} Cross-linguistically, lexical extensions from eye to emotional and dispositional concepts are often encountered. Twenty languages of the sample associate eye with concepts such as ‘jealousy,’ ‘desire,’ ‘hate,’ or the proverbial ‘evil eye.’11 Since only eleven of these languages extend eye to mean ‘vision,’ the evidence does not permit an analysis of these mappings as chained metonymies.

3.2.2 The instrument for action metonymy The body parts arm, finger, foot, and hand are frequently extended onto lexic . The languages that extend ear to ‘hearsay’ are Chechen, Danish, Hausa, Kristang, and Turkish. 10. The languages that extend eye to ‘attention’ are Busa, Chechen, Danish, Finnish, Greek, Guaraní, Hani, Hausa, Ilocano, Kurdish, Kyaka Enga, Rendille, Spanish, and Yogad. 11. The languages that extend eye to emotional and dispositional concepts are Basque, Bokobaru, Chechen, Ge’ez, Greek, Hani, Ilocano, Iraqw, Kristang, Kyaka Enga, Ma’di, Pahlavi, Sedang, Selepet, Thao, Turkish, Wardaman, Yoruba, and Yugambeh.

87

88

Martin Hilpert

al meanings that denote actions involving them. However, these mappings do not tend to be further extended. By contrast, the body parts mouth and tongue serve as sources for several chained metonymies. Both mouth and tongue are tightly associated with language cross-linguistically (as are lips and teeth, which are not investigated here), and the conceptual association is metonymic in nature (Radden 2004). Since language is a multifaceted phenomenon that comprises both speech, writing, and meaning, it cannot be precisely determined what individual dictionary entries mean by language. In order not to overinterpret the dictionary compilers, the sense ‘language’ has been reduced to the sense ‘speech’ in the present analysis. Thirty-two languages of the sample extend mouth to the lexical concept ‘speech’ through the instrument for action metonymy. Of these, nine languages have secondary targets of various speech acts, such as ‘agreement’ (Basque), ‘gossip’ (Maidu), or ‘exaggeration’ (Efik).12 The relation between speech in general and these speech acts in particular is captured by the means for action metonymy (Radden and Kövecses 1999: 37). Speech is the means to accomplish a wide range of social activities, so the metonymy has a strong experiential motivation. The body part term tongue undergoes exactly the same developments with comparable frequency. Twenty-six languages have the extension of ‘speech,’ and six of these have the additional meanings of different speech acts. A less frequent lexical extension of both mouth and tongue is ‘word,’ which is also dependent on the intermediate step of ‘speech.’ Efik and Sedang derive the concept from mouth, while Inuktitut and Zapotec derive it from tongue. Like the idea of language, the idea of a word is fairly complex. For the present purposes, it is taken to mean a recognizable stretch of speech that is associated with a concept. The mapping of ‘speech’ onto ‘word’ thus instantiates a whole for part metonymy. In (16), the chained metonymies based on mouth and tongue are shown. (16) {mouth, tongue} → speech → {speech act, word}

3.2.3 The container for contained metonymy A cross-linguistically common lexical extension of belly is ‘pregnancy,’ often in expressions that literally mean ‘big belly.’ Of the fifteen languages that have this extension, Basque and Tahitian extend this meaning further to ‘offspring.’ These senses appear to be conceptually dependent; no language without the first extension exhibits the second. The first step in the chain is motivated by the container for 12. The languages that extend mouth to speech act meanings are Balti, Basque, Busa, Efik, Kurdish, Kyaka Enga, Ma’di, Maidu, and Mandarin. The languages that extend tongue to speech act meanings are Basque, Iraqw, Koyukon Athabaskan, Marshallese, Turkish, and Nez Perce. The dictionary of Nez Perce does not list the prior extension to ‘speech.’


contained metonymy, as the womb contains the fetus. Since offspring is the end result of pregnancy, the second step can be viewed as a cause for effect metonymy. The chained metonymy is schematized in (17). (17) belly → pregnancy → offspring

3.2.4 The objects are humans metaphor Cross-linguistically, body parts such as arm, back, belly, face, foot, forehead, and head tend to become extended to refer to object parts. Through such extensions, body part terms are rendered more abstract and often grammaticalize into spatial markers. By contrast, the developments discussed below illustrate how these meanings are mapped onto lexical meanings. Twenty languages of the sample generalize head to ‘top part’ through the objects are humans metaphor. The subsequently targeted lexical concepts ‘beginning’ and ‘end’ are observed in eleven and eight languages, respectively.13 The extension of head to mean any extreme object part regardless of spatial orientation is accomplished through the member of a category for the category metonymy (Radden and Kövecses 1999: 34). This chain is shown in (18). (18) head → top part → {beginning, end} Another lexicalization chain involves the extension from back onto the grammatical meanings ‘back part’ and ‘behind,’ which will be discussed in more detail in section 3.3. Interestingly, five languages extend this chain further to the activity follow.14 This development is intriguing from a theoretical perspective, since it appears to map a grammatical meaning back onto a lexical target. A number of authors (Heine et al. 1991; Lehmann 1995) explicitly reject the notion of degrammaticalization, but several counterexamples to strict unidirectionality challenge strong versions of this position (Campbell 2001: 127). The present example suggests that grammatical forms may indeed spawn off new lexical forms, while the grammatical form itself can stay grammatical. Similar changes can be observed to occur in English (e.g. no ifs, ands, or buts, cf. Heine 2003: 166). However, whether or not these are true instances of degrammaticalization will be left to future debate. The extension from ‘behind’ to ‘follow’ maps a spatial region onto an associated activity through (what may be called) the place for action metonymy. The me13. The languages that extend head to ‘beginning’ are Aitchin, Efik, Hausa, Kristang, Kurdish, Mandarin, Pahlavi, Turkish, and Uzbek. Awa and Balti have this extension despite no indication of the extension ‘top part’ in the respective dictionaries. The languages that extend head to ‘end’ are Aitchin, Finnish, Khwe, Kolami, Kongo, Mandarin, Pahlavi, and Turkish. 14. The languages that extend back to ‘follow’ are Bantawa, Chantyal, Hausa, Kolami, and Lushai.

89

90

Martin Hilpert

tonymy is experientially motivated, as certain places and positions are associated with matching activities. People who lead are in the front, whereas people who follow are behind. Radden and Kövecses (1999: 42) point out that this metonymy is reflected in the English expression I was behind the wheel all day. Of the five languages that exhibit the described chained metonymy, four also have the extension from ‘behind’ to ‘after.’ Despite the close connection of spatial and temporal sequence, the cross-linguistic evidence suggests that these extensions are in fact independent because a number of languages have just one of them. In a similar vein, the extension from back to ‘behind’ leads to the meaning of ‘support’ in seven languages.15 Like the above example, this chain re-lexicalizes a grammaticalized form. The semantic extension from ‘behind ‘to ‘support’ is again motivated through the place for action metonymy. The presence of supporting people from behind in a situation of conflict can be readily extended to the notion of help and support in other contexts. Walsh (1994: 360) identifies this extension in expressions from Murrin-Patha and compares them to English examples such as You have to back your mates, underscoring the wide currency of this sense development. For both this chain and the previous one, a direct mapping of ‘back part’ onto the respective activities would make them ordinary instances of lexicalization, rather than putative examples of degrammaticalization. However, in both cases the deictic position of the followers and supporters seems to be an integral semantic component. In addition, in all languages with these extensions the meaning of ‘behind’ is a conventionalized sense of back, which constitutes independent evidence for the latter view. Thus, we can schematize the chain of conceptual extensions as in (19). (19) back → back part → behind → {follow, support}

3.3 Extensions onto grammatical meanings The extensions of body part terms onto grammatical meanings that were encountered in the database all have in common that they rely on the objects are humans metaphor as a first conceptual mapping. From there on, body part terms are extended to become markers of spatial and temporal relations. Table 2 shows that extensions onto grammatical meanings are encountered less commonly in the database than extensions onto lexical meanings. Only the body parts back, belly, buttocks, face, forehead, hand, and head are found to be extended onto grammatical meanings. In contrast to lexical meanings, most extensions onto grammatical meaning can be shown to consist of several mappings. 15. The languages that extend back to ‘support’ are Danish, Efik, Hausa, Kurdish, Oneida, Pahlavi, and Turkish.


Table 2. The most frequent grammatical extensions from body part terms body part

extension (in number of languages)

back belly buttocks face forehead hand

behind (32), after (12) inside (8) behind (5) in front of (8) in front of (3), before (2) five (6)

A chain of extensions from back to ‘back part’ and further onto the grammatical meaning ‘behind’ is found in thirty-seven languages of the sample.16 This number reflects the generality of this well-known grammaticalization process. Fourteen of those languages extend the meaning of ‘behind’ further to ‘after.’17 Most dictionaries specifically list the extension ‘back part’ or refer to senses such as ‘edge,’ or ‘stern,’ while others just give the entry ‘back.’ Since this English gloss can refer to human beings as well as objects, we cannot count this as sufficient evidence against the prediction of Heine et al. (1991) and Matsumoto (1999), who judge the extension ‘back part’ to be a necessary conceptual step. By analogy with the meaning extensions that were discussed above with example (13) from Ma’di, this chain begins with the image-schematic mapping of a human body part onto a relational object part (Svorou 1994: 90). This mapping can be motivated through the metaphor objects are humans . The chain continues with the part for region metonymy, which maps the object part onto an adjacent region of space. The subsequent mapping onto the temporal domain is achieved through the metaphor time is space. Radden (2006) shows that the conception of time in terms of a spatial front-back orientation is crossculturally very common, but that there are differences with respect to the specific mappings. While most Western cultures map the future onto the front, this is not universally so, as evidenced by the conflicting extensions of face and forehead (see below). A semantic development that unfolds parallel to the above chain starts with the term for buttocks, and proceeds in the same steps as above. This process is not as frequently observed as the extension of back. Only five languages of the sample have the extension from buttocks to ‘behind,’ one of which further extends it to ‘after.’18 16. The languages that extend back to ‘behind’ are Balti, Bantawa, Bokobaru, Busa, Chantyal, Danish, Efik, Ge’ez, Guaraní, Hausa, Ilocano, Inuktitut, Kayardild, Khwe, Koiari, Kolami, Kongo, Kurdish, Kyaka Enga, Lushai, Ma’di, Maidu, Mandarin, Marshallese, Nez Perce, Oneida, Pahlavi, Sedang, Selepet, Tagalog, Thao, Turkish, Wardaman, Yir-Yoront, Yogad, Yoruba, and Zapotec. 17. The languages that extend back to ‘after’ are Busa, Chantyal, Hausa, Khwe, Kolami, Kurdish, Kyaka Enga, Lushai, Ma’di, Marshallese, Nez Perce, Selepet, Thao, and Zapotec. 18. The languages that extend buttocks to ‘behind’ are Aitchin, Danish, Koyukon Athabaskan,

91

92

Martin Hilpert

These two related chains are schematized in (20). (20) {back, buttocks} → back part → behind → after The body part term face develops similarly to back, but the orientation is of course reversed. The development from face to spatial and temporal adpositions is a recognized grammaticalization path. However, the data suggests that it is less common than the extension of back, as proposed by Heine et al. (1991: 130). While twenty languages of the sample extend face to ‘front’ via the objects are humans metaphor, only nine languages further apply the part for region metonymy to derive the spatial meaning ‘in front of.’19 With back, the analogous process was observed in thirty-seven languages. Interestingly, the subsequent mapping via the time is space metaphor in six languages yields the interpretations of either ‘before’ or ‘after.’ While the languages Guaraní and Lushai map the space in front onto anteriority, the reverse happens in Bokobaru, Hiri Motu, Karok, and Ma’di. The body part term forehead also serves as a source for the developments discussed above, albeit less frequently so. Only Ge’ez, Hausa, and Kongo derive a spatial adposition from forehead. All three languages further extend it to a temporal meaning. In Ge’ez and Hausa we find the meaning ‘before,’ whereas Kongo has the meaning ‘after.’ In (21), these meaning developments are summarized. (21) {face, forehead} → front → in front of → {before, after} Body part terms are not only used to indicate directions away from the body, but can also encode containment. An extension from belly to the grammatical meaning ‘inside’ is found in nine languages of the sample.20 Like the extension of back, this change maps a body part term onto a location. It is first extended to denote the ‘inside part’ of not only humans but also objects, and from there on assumes the function of a spatial adposition. Again, the objects are humans metaphor and part for region metonymy motivate the semantic extension. As suggested by Heine et al. (1991: 130) and Svorou (1994: 71), the extension of belly onto a spatial concept is less frequent than the extension of back. In the languages Hausa and Ngizim, belly has further grammaticalized into an inclusive marker, meaning ‘one member within a set.’ This meaning is motivated through the metaphor categories are containers (Lakoff and Johnson 1980). Example (22) illustrates this meaning. Oneida, and Rhade. Koyukon Athabaskan also has the extension to ‘after.’ 19. The languages that extend face to ‘in front of ’ are Balti, Bokobaru, Busa, Guaran, Hiri Motu, Karok, Lushai, Ma’di, and Sedang. 20. The languages that extend belly to ‘inside’ are Guaraní, Hausa, Kanuri, Kayardild, Kurdish, Ma’di, Ngizim, Tohono O’odham, and Zapotec.


(22) gayi da kunu-k mainaucin Ngwajin one from belly-link prince-pl Ngwajin ‘one of the princes of Ngwajin’ (Schuh 1981: 98) belly → inside → inclusive Another extension found in Ngizim gives belly a temporal interpretation, which is rendered by Schuh (1981: 99) as ‘be engaged in.’ The element functions as a temporal preposition that co-occurs with nouns that denote activities, as illustrated in example (23). (23) jàa kunu tlәri 1pl belly battle ‘We were engaged in battle.’ (Schuh 1981: 99) belly → inside → during While found only once in the sample, this mapping is well motivated through the very wide-spread metaphor time is space (Lakoff and Johnson 1980). Heine and Kuteva (2002: 179) report numerous cases of extensions from spatial to temporal containment in different languages, in which a body part is mapped metaphorically onto a temporal region. The chain in (24) schematizes the observed extensions. (24) belly → inside part → inside → {inclusive, during} The cross-linguistically common lexical extensions from belly to ‘emotions’ and ‘character’ seem to be independent from the conceptualization of this body part as a container. Languages such as Rendille or Selepet associate ‘anger’ with the belly, but do not seem to conceptualize this emotion as a contained fluid (cf. Lakoff and Johnson 1980). Basque and Nandi attribute the character trait ‘greed’ to the belly, but lack the extension ‘inside.’ As a final example, twenty languages of the sample generalize head to ‘top part’ through the objects are human beings metaphor. From there on, head takes on the grammatical meaning of ‘over’ through the part for orientation metonymy, but this development is restricted to Finnish, Kurdish, and Ma’di. There is no temporal extension of this spatial concept, as is shown in (25). (25) head → top part → over

3.3 Results and discussion In the introduction it was asked whether systematic differences obtain between serial conceptual mappings that lead to lexical and grammatical meaning respectively. The survey of body part terms shows that most serial extensions begin with one of three conceptual steps, namely, the organ of perception for percep-

93

94

Martin Hilpert

Table 3. Serial extensions onto lexical meanings organ of perception for perception > perception for attention ear (76) → hearing (30) → attention (18) ear (76) → hearing (30) → disregard (4) eye (76) → vision (39) → attention (14) organ of perception for perception > cause for effect ear (76) → hearing (30) → obedience (5) organ of perception for perception > perception for thing perceived ear (76) → hearing (30) → hearsay (5) eye (76) → vision (39) → beauty (3) instrument for action > means for action mouth (76) → speech (32) → speech act (9) tongue (76) → speech (26) → speech act (6) instrument for action > whole for part mouth (76) → speech (32) → word (2) tongue (76) → speech (26) → word (2) objects are humans > member of a category for the category head (76) → top part (20) → {beginning (11), end (8)} objects are humans > part for region > place for action back (76) → back part (39) → behind (37) → {follow (5), support (7)} container for contained > cause for effect belly (76) → pregnancy (15) → offspring (2)

tion metonymy, the instrument for action metonymy or the the objects are humans metaphor. The container for contained metonymy is observed in just one chained metonymy. Subsequently, semantic extensions are motivated by a wider variety of metonymic and metaphorical mappings. The main difference between lexical and grammatical targets in the database concerns their initial conceptual mappings. Table 3 shows that the organ of perception for perception metonymy is a particularly productive first step for serial conceptual extension. Less frequent starting points for lexical extensions are the objects are humans metaphor and the container for contained metonymy. As is well-known, body part terms frequently develop into grammatical markers of spatial relations. Semantically, this development is dependent on the personifying metaphor objects are humans. Table 4 shows that in fact all targeted grammatical meanings make use of this metaphor in a first conceptual step. Of equal importance is the part for region metonymy, which maps the meaning of object parts onto projected areas. The resulting spatial meaning may be further extended metaphorically. The difference between the two tables illustrates how body parts can be conceptualized in two basic ways. On the one hand they metaphorically evoke spatial relations, on the other, they metonymically stand for activities and their associated


Table 4. Serial extensions onto grammatical meanings objects are human beings > part for region > time is space back (76) → back part (39) → behind (37) → after (14) belly (76) → inside part (9) → inside (9) → during (1) face (76) → front part (20) → in front of (9) → {before (2), after (4)} forehead (76) → front part (3) → in front of (3) → {before (2), after (1)} objects are human beings > part for region > categories are containers belly (76) → inside part (9) → inside (9) → inclusive (2) objects are human beings > part for region head (76) → top part (20) → over (3)

concepts. It is worth noting that also the secondary extensions in Table 3 are exclusively metonymic in nature. While the metaphorical conceptualization of body part terms thus commonly leads to grammatical, highly schematic meaning, the metonymic conceptualization accounts for a wide range of more concrete lexical concepts.

4. Conclusion Semantic analyses of body part terms in cognitive linguistics have yielded a plentitude of insights, to which the present analysis makes a small contribution with respect to serial conceptual mappings. While the importance of metaphorical and image-schematic extensions has been widely acknowledged (Johnson 1987; Sweetser 1990), this study presents evidence that metonymy is an important strategy in serial conceptual extension that pertains to the entire continuum from lexical to grammatical meaning. The results of this study confirm earlier observations (Goossens 2002; Hilpert 2006b) that simple extensions are much more frequent than serial ones. In the sample of languages investigated, body part terms such as foot, hand, and heart give rise to a wealth of figurative meanings, but interestingly do not project serial extensions. A possible explanation for the behavior of items like foot is that extended meanings such as ‘walk,’ ‘step,’ or ‘footprints’ still evoke the original lexical meaning, and thereby prevent further extensions. By contrast, an item like eye has extended meanings like ‘attention,’ which are sufficiently general to invite further extensions. With respect to serial conceptual mappings, both Taylor (2002: 342) and Goossens (2002: 367) report that metonymies based on metaphor are rare in their data, which comprises English dictionaries and corpora. The present analysis confirms this finding for conceptual mappings onto lexical targets, but suggests a different view for extensions of body part terms onto grammatical targets. The mapping of

95

96

Martin Hilpert

body part terms onto more schematic, spatio-temporal relations appears to be metaphor-based. Since this semantic development is attested cross-linguistically, it is probably not the purported exception. A characteristic difference between the mappings of body part terms onto lexical and grammatical meanings consists of the preference that the latter show for an initial metaphorical extension. Metaphor is used as a strategy to make body part terms more schematic and generally applicable, which allows for more frequent usage and ultimately leads to grammaticalization. Since a number of grammaticalization processes are clearly not driven by metaphor (Traugott and Dasher 2002), it is worth asking whether other lexical source domains exhibit the same tendencies as body part terms. Serial extensions onto lexical meanings, by contrast, favor mainly metonymic mappings. Unlike metaphor, metonymy does not render the meaning of a body part term more schematic, but allows it to be mapped on a contiguous activity or event. In accordance with Goossens (2002) and Taylor (2002), chained metonymies can thus be viewed as a phenomenon that primarily serves to enrich the lexicon.

References Allan, Keith. 1995. The anthropocentricity of the English word(s) back. Cognitive Linguistics 6: 11–31. Barcelona, Antonio. 2005. The multilevel operation on metonymy in grammar and discourse, with particular attention to metonymic chains. In Cognitive Linguistics: Internal Dynamics and Interdisciplinary Interaction, Francisco Ruiz de Mendoza Ibañez and Sandra Peña Cerval (eds.), 313–52. Berlin: Mouton de Gruyter. Blackings, Mairi J. 2000. Ma’di English – English Ma’di Dictionary. München: Lincom. Bowden, John. 1991. Behind the Preposition: Grammaticalization of Locatives in Oceanic Languages [Pacific Linguistics B–107]. Canberra: Australian National University. Bybee, Joan L., Perkins, Revere D. and Pagliuca, William. 1994. The Evolution of Grammar: Tense, Aspect and Mood in the Languages of the World. Chicago IL: University of Chicago Press. Campbell, Lyle. 2001. What’s wrong with grammaticalization? Language Sciences 23: 113–61. Cyffer, Norbert and Hutchinson, John. 1990. Dictionary of the Kanuri Language. Dordrecht: Foris. Deignan, Alice and Potter, Liz. 2004. A corpus study of metaphors and metonyms in English and Italian. Journal of Pragmatics 36: 1231–1252. Geeraerts, Dirk. 2002. The interaction of metaphor and metonymy in composite expressions. In Metaphor and Metonymy in Comparison and Contrast, René Pörings and Ralf Dirven (eds), 435–65. Berlin: Mouton de Gruyter. Gibbs, Ray W. 1999. Speaking and Thinking with Metonymy. In Metonymy in Language and Thought, Klaus-Uwe Panther and Günter Radden (eds), 61–76. Amsterdam: John Benjamins.


Goossens, Louis. 2002. Metaphtonymy: The interaction of metaphor and metonymy in expressions for linguistic action. In Metaphor and Metonymy in Comparison and Contrast, René Dirven and Ralf Pörings (eds), 349–77. Berlin: Mouton de Gruyter. Grady, Joseph E. 1997. Theories are buildings revisited. Cognitive Linguistics 8: 267–90. Heine, Bernd. 2003. On degrammaticalization. In Historical Linguistics 2001, Barry J. Blake, Kate Burridge and Jo Taylor (eds), 163–79. Amsterdam: John Benjamins. Heine, Bernd, Claudi, Ulrike and Hünnemeyer, Friederike. 1991. Grammaticalization: A Conceptual Framework. Chicago IL: University of Chicago Press. Heine, Bernd and Kuteva, Tania. 2002. World Lexicon of Grammaticalization. Oxford: Oxford University Press. Hilpert, Martin. 2006a. Keeping an eye on the data: Metonymies and their patterns. In Corpusbased Approaches to Metaphor and Metonymy, Anatol Stefanowitsch and Stefan T. Gries (eds), 123–51. Berlin: Mouton de Gruyter. Hilpert, Martin. 2006b. Chained metonymies. In Empirical and Experimental Methods in Cognitive/Functional Research, John Newman and Sally Rice (eds). Stanford CA: CSLI. Hollenbach, Barbara E. 1995. Semantic and syntactic extensions of body-part terms in Mixtecan: The case of ‘face’ and ‘foot’. International Journal of American Linguistics 61(2): 168–90. Hopper, Paul. 1987. Emergent grammar. BLS 13: 139–57. Johnson, Mark. 1987. The Body in the Mind. Chicago IL: University of Chicago Press. Kilian-Hatz, Christa. 2003. Khwe Dictionary. Köln: Rüdiger Köppe. Lakoff, George and Johnson, Mark. 1980. Metaphors We Live By. Chicago IL: University of Chicago Press. Lakoff, George and Johnson, Mark. 1999. Philosophy in the Flesh: The Embodied Mind and its Challenge to Western Thought. New York NY: Basic Books. Langacker, Ronald W. 1987. Foundations of Cognitive Grammar. Vol. I. Theoretical Prerequisites. Stanford CA: Stanford University Press. Lehmann, Christian. 1995. Thoughts on Grammaticalization. München: Lincom. Leslau, Wolf. 1991. A Comparative Dictionary of Ge’ez (Classical Ethiopic). Wiesbaden: Harrassowitz. Levinson, Steven C. 2000. Presumptive Meanings: The Theory of Generalized Conversational Implicature. Cambridge MA: MIT Press. MacLaury, Robert E. 1989. Zapotec body-part locatives: Prototypes and metaphoric extensions. International Journal of American Linguistics 61: 119–54. Matsumoto, Yo. 1999. On the extension of body-part nouns to object-part nouns and spatial adpositions. In Cognition and Function in Language, Barbara Fox, Dan Jurafsky and Laura A. Michaelis (eds), 15–28. Stanford CA: CSLI. Merlan, Francesca C. 1994. A Grammar of Wardaman. Berlin: Mouton de Gruyter. Nerlich, Brigitte and Clarke, David. 2001. Serial metonymy: A study of reference-based polysemisation. Journal of Historical Pragmatics 2: 245–72. Niemeier, Susanne. 2000. Straight from the heart — metonymic and metaphorical explorations. In Metaphor and Metonymy at the Crossroads: A Cognitive Perspective, Antonio Barcelona (ed.), 195–213. Berlin: Mouton de Gruyter. Panther, Klaus-Uwe. 2005. Metonymic reasoning inside and outside language. In LACUS Forum XXXI: Interconnections, Adam Makkai, William J. Sullivan, and Arle R. Lommel (eds), 13–32. Houston TX: LACUS.

97

98

Martin Hilpert

Panther, Klaus-Uwe and Thornburg, Linda (eds). 2003. Metonymy and Pragmatic Inferencing. Amsterdam: John Benjamins. Panther, Klaus-Uwe and Thornburg, Linda. 2004. The Role of Conceptual Metonymy in Meaning Construction. Metaphorik.de 6:2004. Radden, Günter. 2004. The metonymic folk model of language. In Imagery in Language: Festschrift in Honour of Professor Ronald W. Langacker, Barbara Lewandowska-Tomaszczyk and Alina Kwiatkowska (eds), 543–65. Frankfurt: Lang. Radden, Günter. 2006. The metaphor time is space across languages. In Metonymy – Metaphor Collage, Elżbieta Górska and Günter Radden (eds), 99–120. Warsaw: Warsaw University Press. Radden, Günter and Kövecses, Zoltán. 1999. Towards a theory of metonymy. In Metonymy in Language and Thought, Klaus-Uwe Panther and Günter Radden (eds), 17–59. Amsterdam: John Benjamins. Reddy, Michael J. 1979. The conduit metaphor: A case of frame conflict in our language about language. In Metaphor and Thought, Andrew Ortony (ed.), 164–201. Cambridge: Cambridge University Press. Ruiz de Mendoza Ibañez, Francisco J. and Diéz Velasco, Olga I. 2002. Patterns of conceptual interaction. In Metaphor and Metonymy in Comparison and Contrast, René Pörings and Ralf Dirven (eds), 501–46. Berlin: Mouton de Gruyter. Schuh, Russell G. 1981. A dictionary of Ngizim. Berkeley CA: University of California Press. Svorou, Soteria. 1994. The Grammar of Space. Amsterdam: John Benjamins. Sweetser, Eve. 1990. From Etymology to Pragmatics. Cambridge: Cambridge University Press. Taylor, John. 2002. Category extension by metonymy and metaphor. In Metaphor and Metonymy in Comparison and Contrast, René Pörings and Ralf Dirven (eds), 323–48. Berlin: Mouton de Gruyter. Traugott, Elizabeth C. and Dasher, Richard B. 2002. Regularity in Semantic Change. Cambridge: Cambridge University Press. Walsh, Michael. 1994. Body parts in Murrinh-Patha: Incorporation, grammar, and metaphor. In The Grammar of Inalienability: A Typological Perspective on Body Part Terms and the Part–Whole Relation, Hilary Chapell and William McGregor (eds), 327–80. Berlin: Mouton de Gruyter. Wilkins, David. 1993. From part to person: Natural tendencies of semantic change and the search for cognates. Working paper No. 23. Cognitive Anthropology Research Group. Nijmegen: Max Planck Institute for Psycholinguistics.

chapter 5

Arguing the case against coercion Debra Ziegeler National Coalition of Independent Scholars

Recent work in cognitive linguistics has hired the term coercion which was formerly associated with computational linguistics to apply to a number of instances in the study of natural language in which there is an incongruity between the semantics of a syntactic frame and the semantics of lexical items found in it. Some of these instances illustrate areas which could well be described as extensions of aspectual boundaries in which the harmony between lexical and grammatical aspect has been penetrated; others illustrate mismatch in nominal phrases, in which the usual parameters that apply have been violated. The purpose of the present chapter is to analyse three individual cases as they have been presented in recent accounts, and to investigate whether such cases could be explained either by pragmatic factors such as metonymic inferencing, or by considering them in the light of their diachronic development. It will then be questioned whether or not a unique phenomenon of coercion can be justified in the context of natural language at all. Keywords: construction-based grammars, indefinite articles, metonymy, mismatch, progressive aspect, subject-complement coercion, type-shifting

Introduction Recent studies in the syntax–semantics interface, and also in areas of research broadly known as unification grammars (e.g. Michaelis 2003, 2004), have isolated a unique phenomenon to account for apparent inconsistencies of grammatical form and function, known as coercion. Although the term seems to have had its origins in artificial intelligence (Levinson 2000: 246), it is increasingly being discussed in relation to natural language and its usage. In particular, it seems to have increased its application to almost any type of inexplicable incongruity between what is de * The preliminary work on this chapter has benefited from the input of a number of colleagues, including David Denison, Martina Faller, John Payne, and the audience of the Language Sandwidge postgraduate students’ seminar series at the University of Manchester, May 17, 2005. I am grateful also to Elizabeth Traugott for many valuable and constructive comments on earlier versions of this chapter, and to the useful suggestions and criticisms of two anonymous referees for the present volume. A longer, more detailed version has been accepted for Journal of Pragmatics (to appear).

100 Debra Ziegeler

manded or anticipated in the syntactic environment of a structure and what is actually found to be possible in language use. The recent upsurge of interest in such a phenomenon and its increasingly wide-ranging attribution to a diversity of linguistic features lead one to believe that the application of the term as an explanatory device may require a more succinct definition than it is presently afforded. The fact that the phenomenon first emerged in the computational domains of linguistic research makes it necessary to consider investigating it further in natural language usage. The numbers of questions which must arise, then, are many. Issues include the actual use by participants in discourse, and how it contributes to the construction of meaning. Another question which arises is that of the actual processes involved in coercing, and why they should be different from any form of pragmatic reasoning; further puzzles emerge as to when coercion ceases to be coercion - some of the examples provided in the literature, such as those involving the aspectual category of semelfactives (e.g. Piñango 2003), seem to be almost conventionalised to the point of no longer appearing to be coerced in any way. Furthermore, the developmental aspect of coercion is not taken into account in any of the studies so far: can coercion be considered a by-product or a contributor of grammatical change in language? A cursory glance at some of the examples of coercion in the recent literature (e.g. Pustejovsky 1995; Jackendoff 1997) creates the impression, finally, that, in the face of a huge pragmatics enterprise, the introduction of such a notion may be rendered superfluous and that it could well be accommodated within what is already known about historical change or metonymic extension in general. Although many of these questions may require a more intensive study to investigate, the present chapter examines some of the disparate range of examples mentioned in the literature, and offers possible alternative explanations in terms of either metonymy or historical development, or both. The question whether the introduction of a new concept such as coercion enables a better understanding of the construction of meaning, or whether the same type of meaning construction can be obtained within existing pragmatic parameters, will also be addressed in the sections to follow.

1. Defining terms 1.1 Mismatch, coercion and type-shifting The terms ‘mismatch’ and ‘type-shifting’ have also become associated with the topic of coercion in general, and it is necessary to distinguish such concepts within the existing theoretical framework. Much of the recent work on coercion has arisen as the result of extended theories of constructions as the minimal linguistic units from which grammatical analysis may proceed; e.g., the introduction of terms such as

Arguing the case against coercion 101

the Plural construction (Michaelis 2003) to replace earlier categorial distinctions referring to morphological classes only. Mismatch has thus been used to refer to the situation of conflict between the constructional syntax and the lexical filler, or form-function mappings (Francis and Michaelis 2003: 2, cited by Traugott 2006), often creating situations parallel to those that may arise in, for example, the grammaticalisation of a morphological category to new functional domains. In section 2, more specific definitions are provided; for the moment it is sufficient to suggest that coercion is essentially the resolution of mismatch. Thus the form-function incongruity expressed in, for example, Did you eat a pudding? (Michaelis 2003: 270), has been labelled mismatch between the Indefinite Determination construction and the lexical items with which it may prototypically combine; i.e. count nouns. In order to resolve such mismatches, a principle known as the Override Principle is called upon (Michaelis 2003: 268), which determines that if the lexical item is incongruent with the syntactic environment, it is the meaning of the syntactic environment of the construction that wins out over the lexical meaning with which it is combined; i.e. pudding in the above example will take on count-noun status. This is known as Implicit type-shifting and is associated with coercion effects. Explicit type-shifting, on the other hand, shifts the designation of the lexical item when the construction type is conventionally associated with that item (Michaelis 2003: 272), so that the type-shifting is derived from unification of constructions with their arguments; e.g. the Partitive Construction a piece of bread, explicitly shifts the mass noun type (bread) to be construed as a bounded part, via unification with a construction referring to a bounded entity, but the mismatch is still between the construction and its open-class argument (Traugott 2006). The latter is not usually associated with coercion, because the association between the construction type and the lexical argument is conventional.

1.2 Coercion and metonymy In the argument that follows, it will be demonstrated that metonymy may play an important role in explaining the architecture so described in the coercion literature. In artificial intelligence, coercion has often been linked to metonymy, and metonymy has been associated with coercion when constraints are imposed on predicate arguments and the constraints are violated - metonymic resolution seems to . In an example of the transitive construction in English, Traugott (2006) illustrates that coercion is related to the way in which syntactic structure may attract by analogy new component parts. . The level of incongruity may, of course, vary from speaker to speaker. Evidence for such variation may provide justification for the gradualness of degrees of conventionalisation of such incongruities.

102 Debra Ziegeler

serve as the means by which coercion is invoked (Hobbs, Stickel, Appelt and Martin 1993). Thus, in such labelling, there is essentially no difference between metonymy and coercion; however, in the area of aspectual coercion discussed below, the relationship between metonymy and coercion is not so clear-cut, and does not always involve just predicate arguments. Jackendoff (1997) discusses examples of coercion as enriched composition, which also include the subcategory of reference transfer expressions, essentially, referential metonymies, as in The ham sandwich is waiting for his check (from Nunberg 1978). In his account, then, there is little difference between metonymy and coercion. The transferred reference of Jackendoff ’s (1997) examples is only one facet of an entire research field of metonymy. Kövecses and Radden (1998), and Hopper and Traugott (2003) indicate that such a definition is too narrow for most purposes in cognitive research, and propose that the relationships existing in metonymy exist between concepts, and are not just related to the lexical forms traditionally associated with referential metonymy (1998: 38). However, even amongst lexical items, the ‘transferred reference’ definition can be demonstrated to be a part of a wider conceptual sphere which would embrace a number of unrelated lexical items of a similar type, e.g., the ham sandwich above might fall into the category of a part for whole metonymy. Kövecses and Radden (1998: 39) maintain that the use of metonymy is not just a matter of simple substitution, and that language users are endeavouring to access one mental entity through another, in using a sophisticated operation. This operation can be related to a type of Gricean Quantity 2 implicature, in which the speaker communicates less than what is inferred from the message (what Levinson (2000: 37) re-labels an I-based implicature — i.e. ‘Minimal specifications get maximally informative or stereotypical interpretations’). If the means of achieving metonymic shifts in natural language is by Gricean Quantity implicatures, then coercion is no different from a form of conversational inferencing. It appears, then, that the term coercion is being extended way beyond its original uses as a simple computational explanation for the processes involved in metonymy. There are other ways besides metonymy of accessing one mental entity through another, including metaphor, but metaphor differs from metonymy in that the relationship between the two mental entities is analogical, while in the case of metonymy, it is contiguous (Hopper and Traugott 2003: 87–8). In Kövecses and Radden’s account, contiguity of meaning in metonymy can be attributed to the location of the entities within an Idealized Cognitive Model (1998: 39), discussed in Lakoff and Johnson (1980), Lakoff (1987), Lakoff and Turner (1989), and Croft (1993). In . Michaelis (2004: 2) also equates ‘enriched composition’ with coercion. . Barcelona (2003a) describes such metonymies as PART FOR PART metonymies, the sub-domain of FOOD standing for the sub-domain of CUSTOMER, both within the domain of RESTAURANT.


such models the source and target concepts belong to a domain of wider knowledge brought to bear upon the situation, i.e. it is a cognitive domain in which specific entities are located in connective relationships. For this reason Kövecses and Radden (1998: 39) develop their own definition of metonymy as “[. . .] a cognitive process in which one conceptual entity, the vehicle, provides mental access to another conceptual entity, the target, within the same domain, or ICM”. Such a definition is designed to be more inclusive than previous ones which discussed only reference transfer between lexical items, or partial structural ellipsis, and refers instead to the entire reasoning process associated with metonymy. Barcelona (2003a: 245) builds on Kövecses and Radden’s (1998) definition in the following way: “A schematic metonymy is a mapping, within one cognitive domain, of a cognitive (sub)domain, the source, onto another cognitive (sub)domain, the target, so that the target is mentally activated,” preferring the use of the term ‘mapping’ to ‘process,’ ‘source’ to ‘vehicle,’ and ‘mental activation’ to ‘mental access.’ Such definitions allow for non-referential metonymy, such as the possibility of metonymy as a device linking the concepts produced in indirect speech acts, e.g. (Kövecses and Radden 1998: 73):

(1) Can you pass the salt?

in which an ability for action metonymy, in the form of a question about one’s ability, stands for a request for the salt to be passed. Such metonymies are in the form of entire constructions, and do not arise solely out of the substitution of one lexical form for another. Stefanowitsch (2003) discusses similar phenomena in relation to cognitive approaches to metonymy. Another type which involves metonymic processes of a non-lexical nature is a predication metonymy, as in the following (Kövecses and Radden 1998: 73):

(2) She was able to pass the driver’s test.

which is described as a potentiality for actuality metonymy in Panther and Thornburg (1999); the assertion of the subject’s ability to pass the test standing for the actual passing of the test (via a Gricean Quantity 2 implicature, an I-implicature of Levinson (2000), or an R-based implicature of Horn (1984)). A similar type which has been isolated as relevant to the historical development of modal verbs is the potentiality for non-actuality metonymy (Ziegeler 2003), as found in the following example:

(3) She had the ability to pass the driver’s test.

in which the inference obtained instead is that the event was not actualised rather than that it was; this is due to the operation of a Gricean Quantity 1 implicature (Levinson’s 2000 and Horn’s 1984 Q-implicature). Again, it is the subject’s ability only that is expressed, but the use of an expression which is more complex

104 Debra Ziegeler

and more prolix than is necessary (when a less complex means is available, e.g. (3)) leads to the interpretation that the content of such an expression represents the limits of the speaker’s knowledge. The resolution of such metonymies is thus interwoven with both the Quantity maxims of Grice (1975) (and his revisionist successors, e.g. Horn 1984 and Levinson 2000). Panther and Thornburg (2003a: 7) note that the relation between the source concept (i.e. the vehicle) and the target concept (i.e. the intended meaning) is defeasible and contingent, thus alluding to the relevant pragmatic processes of conversational implicature. Barcelona (2003b: 98) cites Papafragou (1996) as also suggesting that the interpretation of a metonym often involves pragmatic principles. Barcelona (2003a: 244) notes the ability for metonymies to become socially conventionalised; i.e. a painting by Picasso is referred to as a Picasso. Such conventions are indicative of the pragmatic strengthening of a conversational implicature. Predication metonymies as illustrated in (2–3) do not reflect mismatch or coercion since such effects are associated more restrictively with ‘ungrammatical’ consequences. However, it can be hypothesised from the above discussion that coercion may well be accommodated within a broader theory of metonymy, and can therefore be also discussed in terms of pragmatic inferencing, the mechanics by which metonymic extensions are created. If it can also be seen that a minimal expression (such as the ham sandwich) can stand for a maximal interpretation (such as ‘the ham sandwich orderer’ — the food for customer metonymy of Barcelona (2003a)), then we are invoking an implicature based on Levinson’s (2000) I-heuristic, or Grice’s Quantity 2 maxim, as noted above; similar examples described in the literature as coercion might be explained in the same way. The only distinction perceptible at this stage is that the field in which coercion metonymy operates may often be more limited than that of other metonymies; i.e. to occurrences at the syntax–semantics interface. Whether this distinction is universal to all types of coercion metonymy, though, will be discussed below.

2. Examples of coercion in recent studies De Swart (1998: 360) describes coercion as syntactically and morphologically invisible, and triggered by the need to resolve semantic conflict in contexts which require particular reinterpretation mechanisms to be fully processed. What she means by this is simply that the evidence of coercion is not found in any given grammatical form; it is derived from what the interpreter normally creates from the association of a lexical item in a specific grammatical environment combined with any additional extra-world knowledge that may be resorted to in the process of interpretation. Interpretation, therefore, is not automated, but is calculated on the basis


of additional relevant information brought to bear upon the situation of utterance. Michaelis (2004: 7) assigns coercion to the interpretation of particular construction meanings, and defines it as occurring “[. . .] when the interpreter must reconcile the meaning of a morphosyntactic construction with the meaning of a lexical filler”. Croft (1991) also refers to coercion, suggesting that it occurs when the lexical items of a construction take on the construction semantics, thus proposing a similar definition to that of Michaelis (2003, 2004), while not referring specifically to construction-based grammars. In general, though, the possibility of pragmatic inferencing is not considered in most coercion accounts, which tend to pass over the role of language users in the interpretive process. Obviously, if the term of reference known as coercion is to endure in its natural language application, it must be identifiable as a unique phenomenon distinct from all other possible means of describing the linguistic processes concerned. In (4) are listed three representative cases discussed in recent literature:

(4) a. Nominal coercion (e.g. Michaelis 2003). b. Complement or subject coercion (e.g. Pustejovsky 1995). c. Aspectual coercion triggered by aspectual conflict; e.g. the English progressive (Michaelis 2004).

The phenomena selected in (4) are a small sample of some of the instances discussed in the literature to date. Scope does not permit a full discussion of every type available, but the three selected are interesting for the fact that they represent three completely different instances of coercion, taking place at different syntactic levels and involving different grammatical phenomena. The constructions associated with (4a) may be discussed with regard to type-selecting, while the progressive in (4c) is referred to as a type-shifting example (Michaelis 2003). According to Michaelis (2003: 266), typeselecting constructions express concord relations, that is, they denote the same kind of values as the lexical features of the items with which they are combined (2003: 272), such as boundedness, configuration and number (as in the Indefinite Determination construction). Type-shifting constructions, on the other hand, perform derivations, according to Michaelis (and also Traugott 2006). The type of coercion referred to in (4b), though, is quite different from the other two and will be examined alongside (4a) and (4c) as a means of determining whether there is any commonality to the phenomena described. Each of these cases will be briefly reviewed in turn below and alternative suggestions will be offered by way of explanation. . It is clear from the literature that the tendency has been in most cases to use decontextualised, constructed examples, which in many ways, are self-fulfilling to the objectives of the studies that use them. In the present study, the same examples have been discussed throughout in most cases, since they are the data representing the claims under investigation. In a more intensive study, it would be obviously necessary to use more examples derived from actual usage.

106 Debra Ziegeler

2.1 Nominal coercion In nominal coercion, the most frequent examples cited include discussion of mass nouns co-occurring with indefinite articles and count nouns occurring as bare nouns, unmarked by a determiner, for example (Michaelis 2003):

(5) She had a beer. (6) You have apple on your shirt.

Michaelis (2003: 261) refers to the introduction of a mass noun into the environment of a count noun (in (5)) as seen from the point of view that the incongruity is the result of the misplacing of lexical items. The syntax (i.e. the Indefinite Determination construction) is thus perceived as basic, and prior, and may be considered under such an analysis to ask for new lexical items with which it may be compatible (Michaelis 2004: 22). If this is the case, it may be further questioned why the lexical items are attracted to occur in inappropriate syntactic configurations, when they are incompatible with the syntactic environment in the first place. In (6) the coercion is claimed to be due to the requirement of the Verb Phrase construction for arguments which are maximal categories, i.e. in the case of NPs, that they have reference (to the existence of a quantified entity) or are linked anaphorically to a maximal category (Michaelis 2003: 269). (7), however, shows that bare NP arguments are not restricted to co-occurrence with verb phrases alone:

(7) Top this first layer of apple with more apples slices [sic] arranged across the others like ripples in a pond. www.marthastewart.com/page. jhtml?type=content&id=tvs5030 (www.google.co.uk: 10/5/06)

In other words, they can appear as arguments of prepositions as well. This may not affect the claims made, but Michaelis (2003: 269) also states that the only lexical nouns unmarked for maximality are mass types. However, mass types can also refer (e.g. The apple was all over his shirt), and non-mass types can also be unquantified, e.g. There were apples all over the ground. It may be clearer, perhaps, to suggest that grammatical morphology marking countability on nominals is normally applied to count nouns, except when there is a need to indicate that the entity referred to by the noun is decomposed or degraded to the point at which it is no longer identifiable as a singular entity, or when there is no specific reference. Such instances may therefore omit the article indicating individuation and be interpreted in the same way as mass nouns. . Specificity may be considered to be, under some interpretations, a discourse-level equivalent to lexical boundedness or individuation.


The incompatibility of lexical with syntactic environment is discussed by Michaelis (2003, 2004) as resolved by a procedure in which the meaning of the syntax wins out over the lexical meaning of the item to be coerced into the environment; i.e. the principle referred to in 1.1 as the Override Principle. The coercion effects of mass to count noun coercion are seen to be produced by either of two operators, (implicit) type-shifting operators (e.g. the Plural construction in two beers, where the mass-noun ‘type’ shifts to count noun status in unification with the Plural construction), or type-sensitive operators, e.g. in (5), where the indefinite article is sensitive to co-occurrence with count nouns (see Michaelis 2003: 263). Most of the cases of coercion described are accounted for by such procedures; thus, beer in (5) acquires count-noun status.

2.2 Alternative explanations: grammaticalisation of the indefinite article A grammaticalisation account may propose instead that the indefinite article is generalising to an extended range of uses, and that coercion is just a representation of such extensions. According to Hopper (1987) and Quirk and Wrenn (1969), the indefinite article was rare in Old English, even rarer than the definite article, according to Quirk and Wrenn (1969: 71). The original form of the indefinite article was an, ‘one,’ which shared with the determiner sum ‘a certain’ the function of marking only specific, or referential noun phrases in the discourse (though there were distributional differences between the two forms (Hopper and Martin 1987)). Thus, non-specific indefinites could be left unmarked, as in the following Old English example (Hopper 1987: 143):

(8) Đeodric was Cristen. ‘Theodric was a Christian.’

As suggested by Hopper and Martin (1987), the indefinite article has gradually spread into domains which were formerly marked with zero (or other, non-specific) determiners. This change appears to begin to take place in Middle English, as both zero marking and indefinite articles may be found interchangeably in similar non-specific or non-referential environments in the same text:

(9) a. For ðat it ilimpð ofte ðat godd sant ane man an oðer to helpe. ‘For it often happens that God sends a man to help another man.’

. According to Traugott (2006) all diachronic innovations can be hypothesised to involve mismatch or incongruities initially. . Hopper does not provide the source of this example.

108 Debra Ziegeler

b. ʒif rihtwis mann habbe swo ʒedon te-fore ðe, ʒif ðat holi writ ne wiðseið ðe naht […] ‘If a righteous man has done so before thee, and if that holy Writ does not contradict thee (…).’ (1150–1250) Vices and Virtues, (ed. Holthausen), p. 101. In (9a) the use of ane is glossed as an indefinite article, even though no specific man is being referred to. The same reference to a singular non-specific man is left unmarked in (9b), suggesting that the indefinite article may have been grammaticalising from this time, and used with a certain amount of variability. The movement of the indefinite article into non-referential domains was thus accompanied by a decrease in the referential specificity of the article form and a gradual loss of semantic substance over time, so that in present-day English, it may be used anywhere, even now encroaching into domains formerly reserved for mass nouns (as in (5)). If such examples are the result of diachronic developments, this may be considered an indication of the reduction of the semantic transparency associated with original meanings of marking numerals. However, the meaning rendered by examples such as (5) is that it refers to a measure unit of beer, and if the semantics of marking singularity were completely lost at this stage, we should not be able to derive the count noun semantics at all. The act of coercion is thus based on analogy with conventional uses of the indefinite article, though the interpretation of the coercion may involve different processes.

2.3 Alternative explanations: metonymy 2.3.1 Mass-to-count coercion If mass-to-count coercion were simply a case of generalisation in grammaticalisation, it would also have to be restricted to the use of the indefinite article - thus marking for plurality on a mass noun type (e.g. three beers please) would be considered as a separate case. The Indefinite Determination construction and the Plural constructions are both said to ‘require’ the features of boundedness, configuration, and number (Michaelis 2003: 271) in their lexical arguments. In (5) it could be argued instead that the interpretation is metonymically derived, and involves a substitutive relationship (a beer standing for ‘a glass of beer’ — this could be perhaps labelled a contained for container metonymy — see Kövecses and Radden (1998: 58)). However, it is not always guaranteed that the target form is so readily retrievable, as it is in similar metonymies supplied by Kövecses and Radden (1998: 58), and if it were, the indefinite article should be more productive in such circumstances. Moreover, the following examples (from Jackendoff 1997: 53) could be construed as a little confusing:


(10) a. *He poured three cements today. b. *There’s some vitamin pill on the counter for you. Jackendoff (1997) discusses count-mass coercion of nouns as examples of enriched composition, claiming that it is not always possible in some cases, as shown in (10), and suggesting that there may be selection restrictions on what is possible and what is not possible in coercive operations. The reasons suggested are that in such novel cases, as in (10b)), it is not entirely clear what the substance in the composition of vitamin pills actually refers to, since there is generally no extra-linguistic world knowledge that can be brought to bear upon the situation. Similarly, (10a) would only have relevance, perhaps, for someone working in the building trade. Perhaps such examples might fall outside of what Barcelona (2003a) has referred to as socially conventionalised metonymies, as noted above, and if it is not possible to conventionalise them, they may not be acceptable as coercions. Similar examples are discussed by Ruiz de Mendoza Ibáñez and Pérez Hernández (2001) as material for entity containing/holding the material metonymies, as in I would like three butters please which they suggest to indicate ‘three slices (of toast, etc.) with butter’. (This would be open to construal too, though, as at first glance three butters might instead be construed to mean ‘three pats of butter.’) Thus it is considered that it is not a case of generalisation in grammaticalisation which is motivating such uses; it is hypothesised that appearance in particular circumstances may motivate instead the use of a metonymic device in which the source, a beer, stands for the target, ‘a glass of beer’ or ‘a minimal measure of beer’ (depending on what is considered ‘minimal’ in a given set of circumstances) and that the readily-recoverable target of the metonymy is related to frequency of use or situation (social conventionalisation) rather than to any degree of coercion: the examples are simply examples of transferred conceptual reference in the form of developing metonymies. The overall outcome of these metonymies is an analogy with compatible forms in such contexts, i.e. count nouns, creating the appearance of a grammatically-based metaphor; the process by which the analogy is obtained, though, is retrievable as a metonymy, created by the exploitation of Quantity 2 (Levinson’s (2000) I-inferences), i.e. invited inferences.

2.3.2 Count-to-mass coercion In the case of the count-to-mass operation though, we may have more reason to recourse to grammaticalisation as an explanation. As seen in (8–9), the use of the indefinite article with non-specific referents was not obligatory in earlier stages of English and would have been an added feature, to be placed before a noun as required, when the speaker wished to express singularity as against specificity; that is, non-specific singular count nouns were conceptually comparable to mass nouns.

110 Debra Ziegeler

Unmarked count nouns which are used in a mass sense, then, would never have required marking for individuation. The noun apple in environments such as (6) could therefore have always remained unmarked since no quantification need be expressed (it is unimportant how many apples were involved); there is no specific reference required. If any coercion need be invoked at all, it might only be perceived in the absence of the indefinite determiner some, which also expanded its range of functions to introduce non-specific NPs as well as specific. While it could not be claimed that the historical changes that took place during this period have any bearing on today’s uses of the indefinite article, it can be seen that its original function was to quantify, and thus it applied to count nouns only. Langacker (2002: 72) refers to the ability for count nouns to be replicable as distinguishing them from mass nouns, and suggests that in circumstances in which the boundaries of the replicable noun are irrelevant or cannot be perceived, any count noun can be construed as a mass noun. Amongst the examples he provides are: I don’t like shelf — I’d rather eat table, said by one termite to another, for whom the boundaries of the shelf or the table fall outside the termite’s domain of experience (2002: 78). However, this explains why count nouns can be construed as mass nouns, not how. In the case of (6–7), we cannot claim without an intensive survey of the use of apple since Old English times that the noun had continuously remained unmarked whenever boundary quantification was not required, down through the history of its use until present-day English. However, we can claim that the presentday use of the indefinite article is affected by its earlier associations with marking singular quantity, and therefore retains the inferences of boundedness that Langacker discusses, as an additional meaning besides its grammatical function of marking indefiniteness on count nouns. That is, the indefinite article selects a singular, bounded, portion from the otherwise unbounded mass representing the nominal conception of the generic world of ‘apple.’ As a marker of boundedness, the interpretation produced when it is not used is that the noun is unbounded and unquantified. Kövecses and Radden (1998: 51), however, describe such instances as metonymies, suggesting that they are the result of the use of the constitution ICM, in which substances are conceived of as parts making up whole physical objects, and resulting in the switch between construals of bounded entities and unbounded ones, for example: (11)

There was rat all over the road.

This is described as an object for material constituting that object type of metonymy; i.e., the object rat has been metonymically shifted to refer to rat substance or remains, the material constituting a rat (Ruiz de Mendoza Ibáñez and Pérez Hernández (2001) also agree on this metonymic description for similar ex-

Arguing the case against coercion

amples). The apparent semantic conflict is perhaps due only to the fact that bounded, replicable entities are most frequently used with such determiners, and that when they are not, they derive a pragmatic inference Quantity 1 from the absence of count noun determiners of replicability, suggesting that they are unbounded or non-replicable, as noted above, while at the same time the object for material constituting that object metonymy operates at the lexical level of referential metonymy, and refers directly to the shift from count noun (object) status to mass noun (material) status, albeit with the grammatical consequence of zero determination. It could be argued that referential metonymies can be derived independently of their syntactic environment, though in some circumstances there are grammatical effects, and this is what appears as coercion. It could also be argued that the loss of determination is due to the loss of the specific or referential nature of the rat material, otherwise the determiner some would be used instead. The same bare NP may occur with non-specific reference to an unbounded quantity of count noun objects (as mentioned above); e.g. There were apples all over the road, the unbounded quantity rendering the reference pragmatically non-specific. Perhaps, then, as in Old English, the bare NP marks non-specific reference, but in today’s English only when the NP is not a singular count noun (indefinite articles having long since grammaticalised to mark non-specific, as well as specific, singular count nouns). Thus in both the count-to-mass coercion and the mass-to-count coercion the construction of new meaning can be seen to result from the exploitation of invited inferences of Quantity: in mass-to-count noun coercion, the metonymic adjustments of the source form (e.g. a beer) construct the additional meaning (e.g. a glass/pint of beer), creating new uses for mass nouns via Quantity 2 inferences, while in countto-mass coercion, metonymy associated with Quantity 1 inferences creates the construction of mass construals of count nouns due to the absence of determiners indicating specificity and boundedness. In the following section, it will be seen how extensions of meaning are constructed through the coercion of complement clauses.

2.4 Complement and subject coercion As with coercion of mass nouns to count noun status, the coercion of complements is clearly also a matter of simple metonymic inferencing. The coercion of verb complements has been associated with what is known as the qualia structure of noun phrases, as described by Pustejovsky (1995). Qualia structure, on the surface, appears like a type of argument structure for noun phrases, and involves the consideration of certain aspects of inherent semantic knowledge which are encyclopaedically associated with nouns; e.g. for a noun referring to an object, its purpose or use, its formal structure and the method by which such an object comes into being. Pustejovsky (1995: 77) also maintains that the qualia structure contributes to oper-

111

112

Debra Ziegeler

ations such as type-shifting. In (12), Pustejovsky has supplied a means of recovering the ellipted material: (12) John began the book in which the so-called qualia structure corresponds to extra-linguistic knowledge about the function and nature of the source (vehicle) of the metonymy, the book; i.e. books are for reading or writing, so what John does with the book is in accord with what is found in the qualia structure. Ruiz de Mendoza Ibáñez and Pérez Hernández (2001: 340) also discuss similar examples to (12), suggesting that, rather than examples of ‘enriched composition,’ as described by Jackendoff (1997), they are better described as examples of contextual parametrization of the unrealised generic value of the generic selection restriction provided by the verb. Although Ruiz de Mendoza Ibáñez and Pérez Hernández are not clear on the meaning of generic in this use, they do invoke the presence of a metonymy in such cases, without referring to any particular type of ‘qualia structure’ of the kind referred to by Pustejovsky; it is referred to by them as the object for an action in which the object is involved metonymy. Such labelling focuses on the representation of an action or an event as an entity in itself, in that it can be readily substituted with a more concrete entity. However, they also suggest that context plays an important role in the parametrization process; if the usual activities of John in (12), for example, were to bind books, rather than reading or writing them, the ellipted material would not be as prototypical as could be assumed at first glance. Perhaps this aspect of ‘metonymic reconstruction’ associated with the qualia structure of Pustejovsky (1995: 199–200) can be otherwise described as reflecting the content of an ICM and that it would not be used if there were no extra-linguistic world knowledge available for its retrieval in the first place. Pustejovsky (1995) also discusses the coercion of subjects, as in (13): (13) a. Books bore me. b. The movie frightened Mary. to which another example may be added: c. The food delighted Mary. (Schlesinger 1992: 326) This type of metonymy is not detailed in Kövecses and Radden’s (1998) typology, though it might well be added, since many of those listed in that account are structural in nature rather than lexical. Pustejovsky (1995) indicates again that the qualia structure associated with movies (i.e. ‘watching movies’) and books (‘reading books’) enables a metonymic reconstruction to take place, in which the action of watching movies or reading books has the effect that it does have on the experiencer objects in (13). Similarly, it could be suggested that the qualia structure associated typically with food in (13c) (‘eating the food’) enables the reconstruction of a


similar metonymy. Kövecses and Radden (1998: 54) discuss the use of the instrument for action metonymy in examples such as to shampoo one’s hair. In such examples the instrument is standing for the action verb itself, while in (13) the logical object (e.g. books) of the implied experiencing event (e.g. reading books) is the surface subject (Pustejovsky 1995). In such cases, though, it could be said that the instrument is equivalent to the stimulus of the experiencing event, and the action is equivalent to the experiencing event; the stimulus is therefore standing for the entire experiencing event; thus a parallel metonymy of the type: stimulus for experiencing event could be proposed to account for such shifts. One does not need to resort to the additional mechanics of coercion operations to explain the pragmatic relations that are underlying such metonymies. One perceivable difference, however, between complement/subject coercion and the NP types discussed above is that the level at which the metonymic reconstruction can take place is different is either case. In the nominal types (2.1), the syntactic level remains the same, the reconstruction of missing meaning taking place within the NP itself. In the complement or subject coercion, the metonymic reconstruction of meaning takes place across syntactic boundaries, so that a nominal entity may stand for an entire (non-finite) clause. In this way, the metonymic adjustments illustrated by coercion examples are seen to cover a wide range of grammatical phenomena, and thus may be defined in different terms in each case. These differences will be discussed in section 3. New meanings are constructed in such examples by the adjustments in the semantic roles represented in the subject NPs. With particular reference to examples such as (13c), Schlesinger (1992) found in one study that there were differences in the amount of agentivity accorded to the subjects of the verbs in comparison to the objects of the verbs by native speaker informants; i.e. the subjects were assigned with more agentive senses of control and cause than the objects, even when they referred to an inanimate entity. In the same way the subjects of (13a–b) could be hypothesised to take on the prototypical semantics of subjects, which are causative or agentive. Thus the exploitation of a metonymy in such examples may contribute to the construction of new meaning in the stimulus subjects. In (12), the object NP is less likely to be affected in this way; though in such examples, the semantics of the verb are enriched to account for the qualia structure surrounding the use of the object NP. Meaning is therefore constructed metonymically on the basis of the accessibility of the target concepts in each case, accessed through what Pustejovsky terms qualia structure.

2.5 Aspectual coercion due to aspectual conflict The other type of aspectual coercion noted by Michaelis (2004) and De Swart (1998) was the type of aspectual coercion found in cases in which there is either a clash of

113

114 Debra Ziegeler

aspectual types (labelled by Michaelis (2004) as ‘endocentric,’ i.e. coercion relating to heads of phrases, as opposed to the ‘exocentric’ nature of frame adverbial aspectual coercion, which relates to non-heads). In aspectual conflict either a perfective lexical aspect (such as is associated with verbs expressing events, or Vendler’s (1967) Achievements and Accomplishments) is used in an imperfective grammatical aspectual environment (e.g. progressives or habituals in English), or an imperfective lexical aspect (verbs referring to processes or states) is coerced into an environment in which the central aspectual tendencies are perfective (as in the use of the past tense in English). Logically, a lexical aspectual category will be attracted to a compatible grammatical aspectual category in the first instance; this has been shown for studies of child language acquisition (Anderson 1991), and can be accounted for by a Principle of Relevance (Bybee 1985) whereby developing grammatical morphemes become first attracted to lexical environments which are relevant to the meanings of the morphemes. English does not distinguish grammatical perfectivity from imperfectivity morphologically; i.e. there is no one-to-one correspondence between lexical and grammatical aspect, the use of the past tense being a stage of grammaticalisation further than that of grammatical perfectives (see Bybee et al. 1994), and grammatical imperfectivity expressed using the Progressive may co-occur with lexically perfective types. Thus grammatical aspectual divisions are less clear-cut than in other languages (such as the Slavic and Chinese languages) where this is concerned (see, e.g. Comrie 1976). However, the consequences of aspectual conflict are seen in the creation of modal inferences, such as counterfactuality (see Ziegeler 1995, 2000), and in the apparently overgeneralised uses of the progressive aspect found in second-language English dialects (see, e.g. Ho and Platt 1993). An example of aspectual conflict coercion discussed at length by Michaelis (2004) is that of the progressive aspect. The coercion of lexically stative verbs used in the progressive provides much cause for concern, especially since, as noted by Michaelis (2004: 36–7), the effect of the progressive, when used with non-stative verb types, is to produce a state-like grammatical situation. In most examples so far, the coercion is brought about by conflict between the lexical entry and its syntactic environment. However, the cooccurrence of states with progressives appears anomalous due to aspectual harmony, rather than aspectual conflict, as shown here: (14) Peter is believing in ghosts these days. where the stative verb, believe, is held to be incompatible with the stativity of the progressive aspect. Thus, if aspectual harmony creates coercion, then it must be the case that aspectual conflict is the preferred situation, as in (15): (15) She was winning the race when she got tripped.


However, both are cases of coercion, according to Michaelis’ (2004) account. It has been said that stative verbs appearing in progressives are not acceptable because the progressive already has a stativising effect on non-stative verbs, and they are therefore redundant (e.g. Vlach 1981: 274). Thus, it is questionable why they represent an instance of coercion. Michaelis accounts for this by suggesting that an operator hold must be added to the causal representation of the input state, yielding an Activity representation, whatever the lexical aspect of the main verb (2004: 37). However, this only describes what happens; it cannot account for what speakers do, nor can it explain why it happens. The basis for a coercion account with some progressives and their lexical input is paradoxical, in the face of other accounts in which the lexical input and the construction are in discord. The progressive obviously requires a participle with an aspectual property of a different type from the stative auxiliary head, i.e. an Activity participle, for the construction to stativise it. There is already semantic conflict in that the participle required is of a different aspectual type from the head. So when a participle of the same aspectual type as the head is introduced into the construction, it becomes questionable why there is still semantic conflict. There could only be a possibility of mismatch, then, if the conflict were at different levels — the stative form of the participle in conflict with the construction’s stativising function (though the function has not even been created until the right form comes along). This is puzzling, not least because Achievements and Accomplishments similarly involve coercion (Michaelis 2004: 17–18), and they are also of aspectual types in conflict with the stativising head verb. However, the coercion involved in the latter cases is not due to the fact that a stative form is in conflict with the expected stativising function, and so the coercion is not created by the (assumed) functional requirements of the construction, but simply because the participle type is not an Activity. Thus, there seem to be two different levels of coercion represented here, one which refers to form-tofunction coercion, and the other which refers to form-to-form coercion. It looks very much as though coercion is involved in one case when the participle is of the same aspectual type as the auxiliary, and in another case when the participle is of a different aspectual type from the auxiliary. The evidence from diachronic investigations of the history of the progressive in English, though, can lend a lot of clarity to this discussion, and it would be an oversight not to take them into consideration There are various accounts of the history of the Progressive construction, some which date the construction from Old English times (e.g. Ziegeler 1999) and others which trace it back only to Middle English (e.g. Bybee, Pagliuca and Perkins 1994). In the Old English construction there existed a form similar to the progressive, the agent-noun, which has been noted in some examples from Old English as ambiguous with the sources of the present-day progressive (e.g. Scheffer 1975). The earliest verb types to co-occur with the progressive in Old English texts have been re-

115

116 Debra Ziegeler

ported to be most often intransitive verbs expressing duration (Traugott 1992), i.e. those referred to as the Activities of Vendler (1967). Denison (1993) found that the Old English form was rarely used with an accusative object, thus excluding aspectual types such as Accomplishments and some Achievements, and when there was an accusative object, the genitive form of the object was always used with the progressive. Stative verbs were also found co-occurring with the progressive, as in the following fifteenth-century example, cited from Mustanoja (1960: 595): (16) a. [. . .] we holden on the Crysten feyth and are bylevyng in Jhesu Cryste. Caxton, Blanchardyne and Eglantine Another example from the same text uses a non-volitional stative verb in the progressive: b. They sayd thre men ansuerd them with grete fere that the paleyce and the ysle was belongynge unto the Kynge of Fryse. Caxton, Blanchardyne and Eglantine The reason that it could have occurred so readily with stative verbs is possibly due to the fact that in its Old English uses, the progressive was often used to refer to situations of a more permanent or generic nature. The function of the participle form, if a continuity with the Old English form is maintained, would thus be adjectival in nature, and attributive of the subject, as a passive participle is today (Ziegeler 1999: 79). This is because the construction appeared to express meanings which were often more time-stable and less verb-like than today’s uses are, i.e. it had a generalised imperfective function. Thus, rather than a grammmaticalisation account, the development of the progressive was seen to emerge as a reanalysis out of an agent noun and along a continuum towards a verb (Ziegeler 1999, in press), in much the same way as is proposed by Haspelmath (1994) for participles crosslinguistically: noun > adjective > participle > verb. The anomaly of states with the present-day progressive may therefore be due more to the possibility that they are recessive, and declining in use relative to increasingly dynamic verb types. The spread of the progressive to more dynamic, perfective verb types, such as Accomplishments and Achievements, was seen to co-occur with the rise in transitive types, especially after 1600 (Ziegeler 1999: 93), and this was shown in small samples from the Helsinki Corpus (which showed a 9% rise in Accomplishments), and a larger sample from Visser (1969–73, which showed a 38% rise in Accomplishments after that time). The sudden rise in Accomplishments co-occurring with the progressive in the beginning of the Early Modern English period indicates that these environments were atypical, and, if any coercion were to be associated with the use of the progressive, it must have been in effect at that time. Historically, however, the progressive was primarily aligned with marking durative aspect on durative verbs,


and the shift towards marking perfective, punctual, or telic verb types must have correlated with a reanalysis of the participle towards the verb-end of the noun-toverb continuum. It could be predicted that the reduction of the auxiliary to an enclitic in some instances accompanies this shift. Coercion, therefore, in such cases, is simply the synchronic recognition of a prolonged and developing diachronic process which involves the increasing accommodation of new lexical environments into an expanding historical pattern of reanalysis.

3. Coercion in the construction of meaning While it can be seen from the above three examples that the cases of coercion illustrated can be pinned down to an explanation derived from the cognitive processes of metonymy or historical change and reanalysis, there are still a number of questions arising from the study. The most important are: i. Are all the examples of coercion metonymy illustrated so far identifiable by their occurrence at the syntax–semantics interface? ii. Does the metonymy resolve mismatch, or create it? iii. How does the study of coercion metonymy contribute to the construction of meaning? In answer to (i), there are many types of metonymy which can be considered independently of coercion, so there is no two-way relationship between coercion and metonymy. The classic referential metonymies of the restaurant ICM provide examples, e.g.: (17) Table One’s movin’ upstairs which could be labelled a table for customer metonymy, in which there is no coercion suggested since the derivation of the metonymy does not involve type-shifting, or any kind of form-function incongruity. Instead it involves a violation of semantic selection restrictions occurring between subject and verb, as tables cannot move on their own. The incongruity, then, is found in anomalous lexical combinations: the intransitive verb move selects for a mobile, and often volitional being as subject. Speakers allow for such incompatibilities because they are able to retrieve a possible target reference of the metonymy, e.g. ‘The customers on Table One are moving upstairs.’ On the other hand, a coercion metonymy such as is illustrated in (5), now repeated in (18), displays incongruities at the syntax–semantics interface because of the . Personal observation, Manchester fish and chips café, 26 November 2004.

117

118

Debra Ziegeler

appearance of grammatical functors marking boundedness, quantity and singularity co-occurring with items that they do not otherwise select for: (18) She had a beer Speakers similarly allow for such incongruities to fit the paradigm simply because they are able to retrieve the target of the metonymy quite readily; e.g. ‘a glass of beer’ or ‘a bottle of beer.’ The process of recovering the target of the metonymy is no different from that of recovering the target of any other referential metonymy, and the pragmatic processes may be no different from those already described, i.e. Gricean Quantity 2 implicatures, or the I-inferences of Levinson (2000). The only difference is that, unlike referential metonymies like (17) which affect only selection restrictions, the expression of the metonymy has consequences for the syntax of the construction it creates. However, it is the metonymic retrieval which permits the co-occurrence of the mass noun with the indefinite article. If it is possible to break down the processes involved in what is believed to be coercion, we may find that it may be triggered in such instances first by Zipf ’s (1949) rule of Speaker’s Economy, the source form a beer standing for a glass of beer, producing the metonymy contained for container, as described above. If the target is not easily retrievable, it is questionable whether the metonymic derivation will take place, as seen above in examples such as (10). However, with time and frequency of use, it is likely that examples such as (18) will conventionalise socially, and their co-occurrence with a mass noun will appear as a grammatical analogy. Eventually, it may be the case that the target is no longer consciously retrievable, as in (19) I had an ice cream at intermission. http://www.granta.com/extracts/2094 (www.google.co.uk; 23/05/06) where the measure unit referring to the portion of ice cream is no longer of any relevance, and any metonymic derivation therefore under-specifies the unit of measure. Examples such as (19) may have started out life as mismatches, but the effects of such incongruities are no longer visible. It is possible that once such mismatch has become semantically ‘bleached’ then coercion has taken place, but while the metonymic reconstruction is still live, there are still pragmatic (I-based) inferences involved in the interpretation — this would mean that coercion may be equated with conventionalisation of metonymy. The noun ice cream now appears polysemous expressing either of two senses according to context: (i) a portion of ice cream and (ii) the substance of ice cream. However, simply placing incongruent construction items in juxtaposition does not imply that they are always coerced into co-occurrence and eventual conventionalisation (as shown in (10)). Thus, question (ii) can be answered, and metonymy often acts to create mismatch which can be resolved by reconstructing the target of the metonymy.


It can also be seen that not all of the examples illustrated so far involve coercion at the syntax–semantics interface, and coercion, if it exists, must take place at different levels. Pustejovsky’s (1995) examples, illustrated in 2.4, demonstrate coercion taking place between a clausal entity and a phrasal one; eg. in (13), repeated as (20): (20) Books bore me. There is no incongruity between the construction and the lexical filler in such examples; books stands for reading books and is quite harmonious with the syntactic demands of the argument slot. The example is simply one of metonymy or enriched composition, and does not create mismatch; the source form simply provides lexical access to an extended meaning. If (20) illustrates coercion, the definition of coercion will therefore need to be expanded to include examples where mismatch is not involved, as well as where it is. The reconstruction, furthermore, takes place across syntactic boundaries and is not simply restricted to the internal syntax of a noun phrase. The final example, the Progressive construction, is another case in point. Any reference to coercion as an online processing strategy must take into account the historical factors wherein the developing aspectual form gradually absorbed nonprototypical lexical aspectual categories such as Achievements and Accomplishments into its expanding paradigm. This was a process of reanalysis taking place over a prolonged time period during which the imperfective (associated mainly with Activities) slowly extended its functions to all aspectual classes of verbs. If such examples were included in a definition of coercion, then they would have to include conventionalised diachronic innovations as well, which are reflected in historically expanding grammatical distributions. It is difficult to determine, on such accounts, exactly when coercion takes place, and whether it is a psychological process activated in processing or simply a present-day recognition of a now conventionalised historical change. The inconsistencies over such definitions present problems for a theory of coercion. In answer to (iii), meaning is constructed, therefore, simply in the enrichment of the semantic composition (to use Jackendoff ’s (1997) terms) of the lexical forms with which the syntactic structures are combined, a process taking place by analogy with existing type-specific semantic constraints. If the syntactic environment lends additional meaning to the arguments contained within it, then items such as beer will be less restricted in terms of their boundedness distinctions in the environment of a count noun, eventually becoming socially conventionalised in the same way as Barcelona (2003a) describes a Picasso to refer to a work of art. However, in the construction of meaning in examples of metonymy, or enriched composition, the limits of constructing meaning depend on the facility with which the target of the metonymy may be retrieved, as examples such as (10) reveal. Once the retrieval is no longer salient to the meaning of the metonymy, the source form acquires an ex-

120 Debra Ziegeler

tended grammatical meaning of boundedness (as in (19)), or additional role-related nuances, as in (20) and (12–13). The construction of meaning in examples such as the development of the English progressive aspect allows punctual aspectual events to be subject to a durative interpretation. This may also be simply a matter of construal, as examples such as She is winning the race can only be reasonably understood with regard to a predictive interpretation; i.e. the event referred to may not actually be in progress at the moment of speaking, but the prediction of it may be uppermost in the speaker’s mind, and so imminent that it is expressed as though it were in progress. The situation of mismatch discussed in the present study leaves open the question, if lexical meaning is enriched through the extended syntactic domains in which it may occur, for example in (18–19), then what happens to the source meaning of the constructions themselves (in this case, the Indefinite Determination construction)? It would seem that the expansion of functions of the developing construction may result in a weakening of the source semantics with which it was originally associated. In the case of the Indefinite Determination construction, there is no loss in the semantics of marking singularity and countability (replicability), though the additional inferences of boundedness are only implied as a pragmatic inference when co-occurrence is with mass nouns. In the case of the developing progressive, the historically prior significance of a general imperfective aspectual function becomes less associated with the progressive as it extends to transitive domains, though new meaning is constructed at the same time in the form of pragmatic enrichment and speaker-subjective functions such as prediction, mentioned above. Further investigation into such questions calls for a more in-depth theoretical analysis of the semantics of mismatch situations in order for them to be resolved.

4. Conclusions It has been demonstrated in a handful of examples above that the notion of coercion is a superfluous explanatory tool in accounting for the resolution of anomalies and mismatches at the syntax–semantics interface, and the fact that its use is not limited to apply to the resolution of mismatch, e.g. in 2.4, leads to the belief that the processes described above are no different from cognitive-pragmatic processes already well-known to most accounts. If there is a need to introduce a new label to describe the processes involved, then it must refer to a concept which can be logically isolated as a unique phenomenon. In the examples discussed in the present study, the term is seen to apply to three different types of situations, two of which can be explained by metonymy, and the other attributable to diachronic development and expansion of grammatical functions. However, coercion cannot describe the pro-


cesses used in interpreting metonymy either, as they involve most often the pragmatics of conversational implicature associated with Grice’s Quantity 2 maxim. As such, the term becomes redundant in natural language domains of usage. The advantage of the alternative explanations provided above is that they do not merely describe a process, but they explain it in terms of the use made by speakers in the construction of meaning. Metonymical adaptation takes place either for the benefit of ease of productive effort on the part of the speaker; to reduce the amount of encoding needed to convey the illocution, or for politeness effects in speech act metonymies in order to encode a confrontational message in a less confrontational manner; (e.g. the use of independent if-clauses (If you will come to order . . .) to encode directives — Panther and Thornburg 2003b). Innovation of new metonymies may sometimes affect the grammatical environment, but the reconstruction of the target of the metonymy permits such mismatches and enables the source form to develop new polysemies according to context. The pragmatic inferencing process of metonymy is associated with two of the three coercion examples discussed, whether involving syntax-semantics mismatch or not, and in the third, the progressive aspect, aspectual mismatch is a by-product of diachronic development and is resolved by historical reanalysis. The evidence thus presented indicates that the challenge remains open to future investigations to isolate a definition of coercion that may be useful to all such examples.

References Andersen, Roger W. 1991. Developmental sequences: The emergence of aspect marking in second language acquisition. In Crosscurrents in Second Language Acquisition and Linguistic Theories, Thom Huebner and Charles A. Ferguson (eds), 305–24. Amsterdam: John Benjamins. Barcelona, Antonio. 2003a. Metonymy in cognitive linguistics: An analysis and a few modest proposals. In Motivation in Language: Studies in Honor of Günter Radden, Hubert Cuyckens, Thomas Berg, René Dirven and Klaus-Uwe Panther (eds), 223–55. Amsterdam: John Benjamins. Barcelona, Antonio. 2003b. The case for a metonymic basis of pragmatic inferencing: Evidence from jokes and funny anecdotes. In Panther and Thornburg (eds), 81–102. Bybee, Joan, 1985. Morphology: A Study of the Relation between Meaning and Form. Amsterdam: John Benjamins. Bybee, Joan, Pagliuca, Williamand Perkins, Revere D. 1994. The Evolution of Grammar: Tense, Aspect and Modality in the Languages of the World. Chicago IL: University of Chicago Press. Comrie, Bernard. 1976. Aspect. Cambridge: Cambridge University Press. Croft, William. 1991. Syntactic Categories and Grammatical Relations. Chicago IL: University of Chicago Press. Croft, William. 1993. The role of domains in the interpretation of metaphors and metonymies. Cognitive Linguistics 4: 335–70.

121

122 Debra Ziegeler

Denison, David. 1993. English Historical Syntax: Verbal Constructions. London: Longman. De Swart, Henrietta. 1998. Aspect shift and coercion. Natural Language and Linguistic Theory 16: 347–85. Francis, Elaine and Michaelis, Laura. 2003. Mismatch: A crucible for linguistic theory. In Francis and Michaelis (eds), 1–27. Francis, Elaine and Michaelis, Laura (eds). 2003. Mismatch: Form–Function Incongruity and the Architecture of Grammar. Stanford CA: CSLI. Grice, H. Paul. 1975. Logic and conversation. In Speech Acts, Peter Cole and Jerry Morgan (eds), 41–58. New York NY: Academic Press. Ho, Mian-Lian and Platt, John. T. 1993. Dynamics of a Contact Continuum: Singaporean English. Oxford: Clarendon Press. Hobbs, Jerry R., Stickel, Mark E., Appelt, Douglas E. and Martin, Paul, 1993. Interpretation as abduction. Artificial Intelligence 63: 69–142. Hopper, Paul J. 1987. Emergent grammar. Proceedings of the13th Annual Meeting of the Berkeley Linguistics Society: 139–57. Hopper, Paul J. and Martin, Janice. 1987. Structuralism and diachrony: The development of the indefinite article in English. In Ramat, Carruba and Bernini (eds), 295–304. Hopper, Paul J. and Traugott, Elizabeth Closs. 2003. Grammaticalization. Cambridge: Cambridge University Press. Horn, Laurence R. 1984. Towards a new taxonomy for pragmatic inference: Q-based and Rbased implicature. In Meaning, Form and Use in Context: Linguistic Applications, Deborah Schiffrin (ed.), 11–42. Washington DC: Georgetown University Press. Jackendoff, Ray, 1997. The Architecture of the Language Faculty. Cambridge MA: MIT Press. Kövecses, Zoltán and Radden, Günter. 1998. Metonymy: Developing a cognitive linguistic view. Cognitive Linguistics 9: 37–77. Lakoff, George. 1987. Women, Fire, and Dangerous Things. Chicago IL: Chicago University Press. Lakoff, George and Johnson, Mark. 1980. Metaphors We Live By. Chicago IL: University of Chicago Press. Lakoff, George and Turner, Mark. 1989. More than Cool Reason: A Field Guide to Poetic Metaphor. Chicago IL: University of Chicago Press. Langacker, Ronald W. 2002. Concept, Image, and Symbol: The Cognitive Basis of Grammar. (2nd ed.). Berlin: Mouton de Gruyter. Levinson, Stephen. 2000. Presumptive Meanings: The Theory of Generalized Conversational Implicature. Cambridge MA: MIT Press. Michaelis, Laura. 2003. Headless constructions and coercion by construction. In Francis and Michaelis (eds), 259–310. Michaelis, Laura. 2004. Type shifting in construction grammar: An integrated approach to aspectual coercion. Cognitive Linguistics 15: 1–67. Mustanoja, Tauno F. 1960. A Middle English Syntax. Helsinki: Société Néophilologique. Nunberg, Geoffrey. 1978. The Pragmatics of Reference. Bloomington IN.: Indiana University Linguistics Club (Reprint of Ph.D. dissertation, 1977, City University of New York). Panther, Klaus-Uwe and Thornburg, Linda. 1999. The potentiality for actuality metonymy in English and Hungarian. In Metonymy in Language and Thought, Klaus-Uwe Panther and Günter Radden (eds), 333–57. Amsterdam: John Benjamins. Panther, Klaus-Uwe and Thornburg, Linda. 2003a. Introduction: On the nature of conceptual metonymy. In Panther and Thornburg (eds), 1–20.


Panther, Klaus-Uwe and Thornburg, Linda. 2003b. Metonymies as natural inference and activation schemas. In Panther and Thornburg (eds), 127–47. Panther, Klaus-Uwe and Linda Thornburg (eds) 2003. Metonymy and Pragmatic Inferencing [Pragmatics and Beyond New Series 113]. Amsterdam: John Benjamins. Papafragou, Anna. 1996. On metonymy. Lingua 99: 169–95. Piñango, Maria Mercedes. 2003. Dynamic properties of enriched composition. In Francis and Michaelis (eds), 311–30. Pustejovsky, James. 1995. The Generative Lexicon. Cambridge MA: MIT Press. Quirk, Randolph and Wrenn, Charles L. 1969. An Old English Grammar. London: Methuen. Ramat, Anna G., Carruba, Onofrio and Bernini, Giuliano (eds). 1987. Papers from the 7th International Conference on Historical Linguistics. Amsterdam: John Benjamins. Ruiz de Mendoza Ibáñez, Francisco and Pérez Hernández, Lorena. 2001. Metonymy and grammar: Motivation, constraints and interaction. Language and Communication 21: 321–57. Scheffer, Johannes. 1975. The Progressive in English. Amsterdam: North Holland. Schlesinger, Izchak M. 1992. The experiencer as an agent. Journal of Memory and Language 31: 315–32. Stefanowitsch, Anatol. 2003. A construction-based approach to indirect speech acts. In Panther and Thornburg (eds), 105–26. Traugott, Elizabeth C. 1992. Syntax. In The Cambridge History of the English Language, Vol. 1, Riachard M. Hogg (ed.), 168–289. Cambridge: Cambridge University Press. Traugott, Elizabeth. 2006. The concepts of constructional mismatch and type-shifting from the perspective of grammaticalization. TS. Vendler, Zeno. 1967. Linguistics in Philosophy. Ithaca NY: Cornell University Press. Visser, Frederikus Th. 1969–73. An Historical Syntax of the English Language. Parts I–III. Leiden: EJ Brill. Vlach, Frank. 1981. The semantics of the progressive. In Tense and Aspect [Syntax and Semantics 14], Philip J. Tedeschi and Annie Zaenan (eds), 271–92. New York NY: Academic Press. Ziegeler, Debra P. 1995. Diachronic factors in the grammaticalization of counterfactual implicatures in Singaporean English. Language Sciences 17: 305–28. Ziegeler, Debra P. 1999. Agentivity and the history of the English progressive. Transactions of the Philological Society 97: 53–101. Ziegeler, Debra P. 2000. Hypothetical Modality: Grammaticalisation in an L2 Dialect. Amsterdam: John Benjamins. Ziegeler, Debra P. 2003. The development of counterfactual implicatures in English: A case of metonymy or M-inference? In Panther and Thornburg (eds), 169–203. Amsterdam: John Benjamins. Ziegeler, Debra P. In press. Interfaces with English Aspect: Diachronic and Empirical Studies [SLCS 82]. Amsterdam: John Benjamins. Zipf, George K. 1949. Human Behavior and the Principle of Least Effort: An Introduction to Human Ecology. New York NY: Hafner.

chapter 6

When Zidane is not simply Zidane, and Bill Gates is not just Bill Gates Some thoughts on the construction of metaphtonymic meanings of proper names Mario Brdar and Rita Brdar-Szabó Josip Juraj Strossmayer University, Osijek/Eötvös Loránd University, Budapest

The present chapter deals with the problem of the construction of meaning of figuratively used personal names in the basic constructional schema Det + Xpersonal name + of Y. After a brief overview of two currently dominant philosophical approaches to problems of reference that are at odds with cognitive linguistic views, we proceed to outline a path along which their meaning is developed in several successive steps. The starting point is the enlistment of our total encyclopedic knowledge about the bearers of proper names organized in complex matrixes of domains. We outline how figurative meanings arise in a step by step fashion, involving tiers of metonymic mappings interspersed with metaphoric mappings. In doing so we also demonstrate that the paragon model can be elaborated in such a way that specific mappings become well motivated. The process of the construction of figurative meaning is shown to be complex, dynamic and flexible. Its output can be revised at every step, further enriched with information, or subsequently depleted of it, depending on the cognitive operations employed. Keywords: active zone, meaning construction, metaphor, metonymy, paragon, proper names

1. Introduction Proper names have always been of great interest to almost anyone concerned with language. A whole range of more or less related issues have been raised and various proposals and/or claims put forward. Some of these issues are quite philosophical in nature, e.g. when it is suggested that names are outside language proper, that they are untranslatable, or even that they are inherently empty of reference, i.e. that they do not have meaning. On the other hand, in more linguistically minded contributions, we also find that discussions of the grammatical structure and behaviour of proper names are interwoven with thoughts on their reference. That this topic is genuinely evergreen can be shown by referring to a whole series of recent linguistic

126 Mario Brdar and Rita Brdar-Szabó

articles dealing with one or the other aspect of proper names, such as Lehrer (1999), Anderson (2003, 2004), Barcelona (2003, 2004), Kleiber (2004), to name just a few. Our starting point will be the fact of the figurative use of proper nouns evident in examples such as those below.

(1) a. Steven has a bag of tricks, a good passer, can operate in confined areas and is the Zidane of Villa whose left foot is nearly as good as his right. b. Sarkozy is described by MEDEF, French CBI, as ‘the Zidane of Finance.’ c. Toni Morrison lives in the ‘Champions League’ of literature. She is the Zidane of language, beguiling, intoxicating and deadly in front of goal. (2) a. He has been called the ‘Bill Gates of Africa.’ b. “He wants to be the Bill Gates of cloning technology.”

The phenomenon is wide spread, the basic constructional schema Det + Xpersonal name + of Y, (or alternatively also Y’s Xpersonal name, if Y happens to denote an area/ location or temporal period) more or less the same cross-linguistically, or varying only slightly depending on the structural resources of the language in question. Barcelona (2003 and 2004) adds to the studies of English, his own discussions of equivalent examples of this construction type in French, Spanish, German, and Italian. By way of example cf. a similar example from German:

(3) Hegel der Bundesrepublik [Headline in a newspaper article] Hegel of Federal Republic Der Gegenwartsklassiker Habermas ist auch ein wenig zum the current-classic Habermas is also a bit into-det Staatsphilosophen geworden, zum Hegel der Bundesrepublik state-philosopher become into Hegel det-gen Federal Republic.of Deutschland. Germany ‘The present-day classic Habermas has turned into somewhat of a state philosopher, into the Hegel of the Federal Republic of Germany.’

In this chapter we focus on a fairly specific phenomenon: the construction of metonymic and metaphoric meanings of proper nouns denoting humans. Specifically, we attempt to identify possible metonymic and metaphoric steps in this process of meaning construction. The chapter is organized as follows. Section 2 is a brief overview of the inadequacies of two objectivist philosophical approaches to the problems of reference that are at odds with cognitive linguistic views. In Section 3 we proceed to outline a path along which the meaning of a proper name is developed in several succes-

When Zidane is not simply Zidane 127

sive steps, involving several layers of metonymic and metaphorical mappings. In particular, we argue for a refinement of paragon as a metonymic model, involving several metonymic tiers and capable of accounting for the axiological dimension of paragons. In the concluding section we list some open questions for further research following from our findings, but also sketch some possible solutions to some of these questions.

2. Objectivist theories of reference vs. cognitive approach to the figurative use of proper names There are two objectivist philosophical theories of meaning that currently enjoy the status of classics, both of which were also intended to account for proper nouns. In Fregean (Frege 1952) tradition (cf. Russell 1956) proper nouns are assumed to be uniquely identifying descriptions. On the other hand, in the Millian (Mill 1843/1949) tradition, more recently represented by Kripke (1972), proper names are considered to be semantically empty. The former has been referred to as the descriptivist view of reference or reference-via-meaning doctrine, while the latter has been dubbed the causal-historical view or direct-reference doctrine. The general inadequacies of both of these theories have been discussed at great length in Lakoff (1987: 168ff) and Lakoff and Johnson (1999: 98ff), but we are concerned here with some special problems that arise when we consider figurative uses of proper names. In order to understand these, let us first sum up the essence of the two theories. As far as proper names are concerned, the descriptivist view of reference relies on two basic assumptions: i. every proper name is associated with a description; ii. an object is the referent of a proper name if and only if it uniquely or best satisfies the description associated with it. The description is uniquely satisfied when it is true of an object and only of it. If, on the other hand, no object entirely satisfies the description, it is assumed that it refers to the unique individual that satisfies most of the description. If the description is not satisfied at all or if many objects appear to satisfy it, it is concluded that the name does not refer. Bearing in mind that reference and (linguistic) meanings are not quite the same phenomena, it is nevertheless easy to see that the descriptivist view of reference is beset with a host of problems when it comes to the figurative use of proper names. It is more than obvious that for either Zidane or Bill Gates, there are cases in which there is no single referent that uniquely or best fits the description (whatever it may be). What is more, some of our examples analysed below, such as (10–11), show that


practically any object may be claimed to appear to satisfy the description. At the same time it is also clear that these names are not semantically empty. According to the causal-historical view, a proper name is introduced into the linguistic community more or less ostensibly. It is an act of pointing that indicates the fixing of its reference by a historical individual at a certain point for the purpose of referring to a given individual, and continues to refer to that individual as long as speakers are linked to the initial referring act by means of a causal chain of successive acts of reference. In other words, a speaker who uses a name acquired it from another speaker who in turn acquired it from another speaker, and so forth, up to the first speaker using that name. The reference of the name remains thus fixed through history. As Lakoff and Johnson (1999: 99) rightly point out, “[i]t is not explained how mere pointing and naming can establish a ‘rigid’ symbol-to-world designation or how that ‘rigid designation’ is able to remain in place over millennia.” But we might also turn the table and ask ourselves: how is that ‘rigid designation’ able to change in order to accommodate figurative uses of proper names? In other words, the doctrine, as it stands with its too stable and rigid designations, does not allow for such figurative uses with a reference that is different from the original. One might claim that such figurative uses of proper names are not linked any longer via a causal chain to their original literal uses but constitute new acts of naming and thus start new causal chains. But even if we adopt this view (although it is clear that the figurative uses are not totally severed from their literal uses), we still face the same problem that we faced before: how is it possible that their meanings arise so to speak ex nihilo? But even worse, how are speakers of a language able to keep record of such cases of ‘homonymy’ and use the names correctly if they do not possess a telepathic ability and log on to a mental network tracking constant shifts in meaning? Considering the range of the combinations of figurative meanings that we claim below to be attached to one and the same name, the burden of keeping such cases of ‘multiple homonymy’ in check might become intolerable. One might also claim that the figurative uses of proper names are unlike their normal uses because the real naming takes place in the subject position. Because in the examples in (1–2) we have figuratively used names in the predicative position, it may be said that these are just ways of assigning properties or attributes to subjects. Again, this does not solve the problem, as many of these names, together with their determiners and modifiers, can also be easily used in the subject position, where allegedly genuine naming takes place. Such analytical pitfalls can be avoided by adopting the cognitive linguistic stance that proper names are not some odd extra-linguistic expressions and that they cannot be without meaning or reference. Further, by assuming that their meanings are not fixed rigidly in advance and outside the context of language use, and that they


can be worked out only by enlisting our total encyclopedic knowledge organized in complex matrices of domains, we are able to show below why it is possible to arrive at the meanings of figuratively used proper nouns at all, and we also suggest some ways in which this is done. The process of such online construction of figurative meaning is bound to be complex, dynamic and flexible, and its output can be revised at every step, further enriched with information, or subsequently depleted of it, depending on the cognitive mechanisms employed, mainly metonymy and metaphor. According to Panther and Thornburg (2004: 113), metonymy (and we may also add metaphor here as well) “not only makes target meanings accessible but also available for further elaboration in discourse.” This is precisely what we intend to demonstrate in this chapter by analyzing a series of metonymic and metaphoric mappings taking place at various tiers, one tier feeding the next. The starting point is, as we pointed out earlier, an ICM-like knowledge base containing our total encyclopedic knowledge about the bearers of individual names, which can be just (barely) familiar or even famous.

3. Metonymy and metaphor in the construction of the meaning of figuratively used proper names 3.1 Tiers of figurative meaning It is relatively easy to see that the figurative use of proper names is often patently metaphtonymic, i.e. it combines both metaphorical and metonymic mappings (cf. Goossens 1990). Let us consider some more examples and see how such figurative meanings are constructed:

(4) a. Yes, I am the Bill Clinton of the blogosphere. b. The Bill Clinton of Hollywood maintains his reputation as Mr. Good Time Genius while wallowing in oceans of slime.

(5) a. Questions for the Mick Jagger of Gardening… b. Actually, I always wanted to be the Mick Jagger of cinema but I think I’m the Keith Richards, by force of habit, perhaps.

(6) a. “I call him the Einstein of horses.” b. The Einsteins of the Deep [the title of an article about dolphins]

The central questions we have to address are how the metonymy and metaphor interact here, and how the two processes bring about the suspension of the inherent grounding of proper names. To begin with, we must make it clear that metonymy is at the base of all these expressions (cf., however, Thurmair 2002, who complete-


ly leaves out metonymy in the analysis of her German examples analogous to our data and analyses her examples solely in terms of metaphor), and that the metaphorical layer follows metonymy. In Goossens’ terms, these expressions count as metonymy within metaphor. Metaphorical mappings are made conspicuous by the of-postmodification. Bill Clinton and Mick Jagger are typically associated with the domains of politics and rock music, respectively. When someone is said to be the Bill Clinton of the blogosphere or of Hollywood, this is a clear case of applying these labels to very different domains. On the other hand, what it means to be the Bill Clinton of the blogosphere or of Hollywood, or the Mick Jagger of gardening or of cinema, can only be worked out if we know how the two names relate to their respective primary domains. These basic pieces of information on them, we claim, are provided by several tiers of metonymic mappings. An analysis of this construction type with proper names has been advanced in Barcelona (2003 and 2004). He accounts for expressions such as: (7)

Lope de Vega was not the Spanish Shakespeare.

by assuming that they are motivated by two metonymies crucially resting on the concept of paragon. According to Barcelona, first a stereotypical model of the proper name’s referent is activated as a paragon, in the course of which we arrive at a characteristic property by means of a metonymic mapping. In this specific example, Shakespeare is “a writer endowed with immense literary talent,” and it is claimed that Shakespeare as a writer endowed with characteristic properties is mapped “onto the rest of our knowledge about him” (Barcelona 2004: 364). An essential element of his Shakespeare network is an ICM-like class or collection of individuals comprising “individuals characterized by one or more of the relation and properties imported from the conceptual model” (Barcelona 2004: 364) created by the above metonymy. In the case of his example given in (7), “the common relation in which all the members of the class participate is having immense literary talent.” Finally, Barcelona postulates a second metonymy, ideal member of the class, connecting the model in which Shakespeare is conceptualized as a writer endowed with immense literary talent as a source domain with the class of individuals with immense literary talent as the target domain. Note that Barcelona does not explicitly mention the metaphorical nature of such expressions in spite of his talking about source and target domains. First of all, it is not immediately obvious that metaphor is involved at all, i.e. it is not so clear whether Spain should count here as a markedly distinct domain. Secondly, his network is designed in such a way as to circumvent any need of metaphorical mappings in this particular example, as the class of individuals with immense literary talent above is a world literature class, i.e. has members from all over the world.

When Zidane is not simply Zidane

3.2 The paragon as a metonymic model While there is no doubt that metonymies are followed in some cases by metaphor in this construction type, our data indicate that there are also some cases where metaphor is completely missing, and that there are also many less than clear-cut cases implying a metonymy-metaphor continuum rather than a sharp dividing line between them. More importantly, we propose on the basis of our data an alternative analysis of the construction with more than just two metonymic tiers. The reasons motivating the refinement we propose have to do with the concept of paragon as a metonymic model, which in its original formulation (cf. Lakoff 1987) is very much like a blackbox. Paragons as metonymic models, just like generators, ideal cases, social stereotypes, etc., are said to operate on the basis of the same single metonymic process, which is called category for defining property by Kövecses and Radden (1998: 54). What we see as the most conspicuous problem with it is that it is not clear where the axiological notion of representing “either an ideal or its opposite” (Lakoff 1987: 87) comes from. We concur with Barcelona that the process of the construction of meaning in the case of such figuratively used proper names is of basically metonymic nature. Metonymy is used here like an extremely precise tool, almost like a conceptual scalpel, to cut out just the right amount of information about the people in question from our knowledge base (on this function of metonymy cf. Brdar and Brdar-Szabó’s 2001 discussion of the interaction of metonymy and metaphor, as well as Buljan and Gradečak-Erdeljić 2003 and Belaj 2005 on the role of metonymy in conceptual blending). However, as pointed out above, we think that Barcelona’s first metonymic tier, the one based on the notion of paragon, is not sufficiently motivated. Specifically, the mapping from Shakespeare to Shakespeare as a writer endowed with immense literary talent can be elaborated so as not to appear so abrupt. The problem may not show up in all the data, specifically it is not conspicuously apparent in the data analysed by Barcelona, as his examples are compatible with the assumption of an ICM-like class or collection of individuals comprising “individuals characterized by one or more of the relation and properties imported from the conceptual model” (2004: 364). The clearer the case for a final tier with metaphorical mappings linking two quite distinct domains, the more difficult it becomes to maintain such a class: it would really be a rag-tag collection of individuals. This is why Barcelona’s analysis and ours are alternative and even complementary. In some cases the paragon model may work as a compressed shortcut model, but if it is analysed, we claim, our additional metonymic tiers will show up. A possibility that should be contemplated is whether perhaps by assuming these additional metonymic tiers we eliminate any need for the other ICM-like class or collection of individuals with a speci-

131


fied property. In other words, the specific properties need not be duplicated in two models. Expressions such as Zidane or Bill Gates in our initial sets of examples are multiple metonymies where some tiers might easily go unnoticed. First of all, Zidane in the examples in (1) must be interpreted as ‘Zidane the footballer,’ i.e. we have a whole for part mapping taking place here. The whole of our encyclopedic knowledge that we have about the person labelled by the proper name referring to a particular real person, i.e. Zidane, must constitute a domain of its own. This claim need not meet with unanimous approval, but note that it is not unusual, as Barcelona (2004) and Ruiz de Mendoza and Mairal (this volume) speak of Shakespeare and Proust, respectively, as domains. This domain must be allowed to intersect with the football domain, as part (if not most) of our knowledge about Zidane has to do with football. ‘Zidane the footballer,’ as the metonymic target, is the intersection of the holistic Zidane domain with this general football domain. Zidane as a whole has many properties, and some of these, in fact the most salient ones, are the subdomain of ‘Zidane the footballer’ within the football domain. When we talk about the whole for part mapping here, we mean that the personal name Zidane as a metonymic vehicle, normally used to cover all our knowledge about Zidane, organized in a single domain or domain matrix, actually comes to mean something much more precise. It is reduced to ‘Zidane the footballer’ as the metonymic target, defocusing or almost totally obliterating other elements of knowledge about him. The highlighted portions of the domain become so prominent as to overwrite the whole structure of the original knowledge base (be it a domain or matrix). This is what we intend to indicate by the arrow. The arrow direction may at first appear confusing: if we talk about the linguistic form as the vehicle, the arrow should of course go in the other direction, indicating that Zidane stands

Zidane

Zidane the footballer

football

Figure 1. Domain intersection as target of entity for active zone metonymy


for something more specific, i.e. ‘Zidane the footballer.’ We use the arrow just the way it is used in Barcelona (2004), which we take as our starting point. A quote from Barcelona (2004: 364–6) may explain this: Box (a) represents the mapping of Shakespeare as endowed with his characteristic properties onto the rest of our knowledge network about him. This mapping results in a stereotypical model of Shakespeare as primarily, in fact exclusively, a writer with immense literary talent. Box (b) represents the mapping of this stererotypical view of Shakespeare onto the class of writers with immense literary talent. Both are metonymic mappings and are symbolized by the arrows.

Of course we are aware that, as Panther (2005: 2) puts it, one domain does not obliterate the other, i.e. both are “still, to some degree, conceptually salient or activated.” One might be tempted to approach ‘Zidane the footballer’ in an alternative way: not as a metonymy in a strict sense, as defined in Panther and Thornburg (2004), but as a process of construal of salience operating on facets as readings within senses, or facetization for short (cf. Paradis 2004). However, what we see as being involved in this metonymic mapping in the Zidane case cannot be considered relationally autonomous senses. They do not stand in hyponymic or hyperonymic relations to other senses that would be incompatible in coordination, they are not compositionally autonomous, as predicates do not apply independently to facets, etc. Facets of meaning call up more than one set of qualia roles, and it is not immediately obvious that this is the case in what Barcelona and we would still be inclined to consider a case of metonymy. This type of metonymy with proper names, briefly mentioned in Barcelona (2004: 360 and 370ff), is discussed in some detail in Barcelona (2003) under the rubric entity for active zone, but is treated separately from examples like (7). The metonymy in question effects the conceptual division of an otherwise unitary domain (Barcelona 2003: 29). We think that the process in question, regardless of whether we consider it some type of whole for part metonymy, or as facetization, should be postulated as basic input for any further metonymic and metaphorical mappings. Let us now demonstrate the need to postulate such a metonymic mapping. It may be considered a “lean” metonymy when compared with Barcelona’s metonymy, i.e. it does not yet map more specific bits of the type “X endowed with immense talent as far as Y is concerned” by contrasting them with some other examples. Against the background of our encyclopedic knowledge of Zidane as a football player we know that his performance during the World Cup 1998 in France was one of the highlights in his career. Knowing that example (8a) is indeed about Zidane and not someone else, we could safely assume that this is a statement in which his extremely good performance is meant, along the lines of Barcelona’s analysis of Shakespeare above.


(8) a. He is one of the players that I would pay any amount of money to watch but he is not the Zidane of 1998.

This also accounts for an interview in October 2004 in which Zidane refers to his best period in Real Madrid in the following way:

(8) b. – If you review your career at the Club, what mark would you give it? – Lately I cannot say that I see myself very well, let say I have been playing normally. But I would choose the Zidane of the first two years. It’s logical because during these years we have won many things and I want to carry on. That’s why I’m here.

At the same time, against this background, an expression like:

(8) c. the Zidane of 2004

cannot possibly be interpreted in the same way. It is a construal in which his performance is average or normal. We could also isolate the worst period in his career so far, full of crises and bad performance, and use the same construction to refer to it. What is more, we can also refer to his future performance, as in:

(8) d. the Zidane of 2010 e. the Zidane of 2020

in which case it is impossible to foresee whether his performance will still be excellent or not. This shows that the assumption that a paragon interpretation is activated right at the start is not founded on solid facts. Consider also the possibility of referring to the Zidane of 2040, when he is not very likely to be playing football professionally. This may be interpreted as being about Zidane’s potential career as a football coach (successful or not), or some other aspects of his life with more or less glorious moments, say his problems with old age, his family situation in the future (perhaps alluding to his future role as a caring and beloved grandfather), etc. Consider finally the following batch of examples highlighting various stages in the life of Ronald Reagan, two of which certainly do not lend themselves to any paragon analysis:

(9) a. The Ronald Reagan he depicts is a man utterly inconsistent in his beliefs and different from his earlier political days. b. And that’s when we got to know the Ronald Reagan he had now become. c. At times, the past and present got confused, as if the speaker could not decide whether the Ronald Reagan he knew still existed.

When Zidane is not simply Zidane

The first two of the Reagan examples, (9a) and (9b), are certainly different, there is no paragon idea at all. And of course, if it were not a famous person, but, say, John Smith of 1999, no one would see any trace of paragon here at all. This means that the interpretation we need for (2a) and (2b) is not necessarily posterior to the paragon interpretation—it can obtain with ordinary names not susceptible to any paragon analysis. The examples in (8) and (9) may alternatively be interpreted as cases of setting up new mental spaces by means of space-building of-phrases, with a projection link between the current or parent space and the newly setup space. But such links do not in any way preclude a metonymic mapping between a whole and its parts. Another possibility would be to treat these examples, and indeed most of our examples in the present chapter, as cases of conceptual integration. Our objective here was quite similar to Barcelona’s (2004: 373)—to demonstrate the metonymic motivation of the construction under investigation, and not to “represent in a detailed elaborate model the web of connections resulting from these special uses” of personal names. We also share with Barcelona the view that metonymy has a crucial role in blending, i.e. that considerable portions of such webs in blends can be unpacked and linearized into a series of metonymies. These come to be presented in a more compressed and holistic form of blending. Metonymic mappings may stop here, but they may also continue. In order to come closer to an appropriate interpretation of expressions like (1) and (2), we must assume further tiers of metonymic mappings that narrow down the information available in the domain intersection in Figure 1 and modify it. First, within this domain intersection, comprising all the information about Zidane that is relevant to conceptualizing him as a footballer, we suppress the majority of elements that are shared by all football players, and highlight those elements that make him different, even search for his unique traits, such as precise passes, scoring from free shots, etc. In other words, we activate the bearer of property for characteristic property metonymy that brings us closer to the “X endowed with immense talent/skill as far as Y is concerned” meaning, but yet not right down to it, as shown in Figure 2. The evidence for this metonymic tier, admittedly indirect, comes from some unusual examples in which metonymies are topped with a metaphorical tier. Consider now the following examples: (10) a. Described, to his slight embarrassment, as the “Michael Jordan of rugby league” Lauitiiti has arrived at Leeds with a reputation of being the most exciting ball-handling forward in the world. b. Humboldt is the Shakespeare of travelers – as much superior in genius to other travellers as Shakespeare to other poets. Note that in these cases, which clearly involve metaphor, we get some contextual clues as to what correspondences between the source and target domain are intend-

135


Zidane


football

Figure 2. Contrastive properties as the target of the bearer of property for characteristic property metonymy narrowing down the interpretation of Zidane as the ultimate metonymic vehicle via zidane the footballer

ed. Shakespeare as a source domain in (10b) is narrowed down prior to metaphoric mappings to a poet superior in genius to other poets. Example (10a) is perhaps even more interesting, it does not specify directly that Michael Jordan played basketball, it only hints at his being the most exciting ball-handling forward in the world via the target domain. Most of the time, however, the intended (or used) correspondences are not explicitly mentioned, but nevertheless the figuratively used proper names are interpreted in that fashion. However, in a number of our examples retrieved from Google in which human proper names are used to metaphorically designate various physical objects, places or concepts, the intended correspondences are mentioned somewhere in the context as a rule. Consider some such examples: (11) a. Indeed, this is the Monica Lewinsky of burgers: oversized, juicy, a little messy, a burger that you know you should resist, but you simply can’t. b. While John Sprung, television executive at Paramount studios said: “This has turned out to be the Monica Lewinsky of digital television. Everybody knows the truth; nobody cares enough to do anything about it.” (12) a. But carrots? They’re just out there, shrieking, “Hi, we’re some carrots! Love us for it!” They never have to prove themselves. They are the Gwyneth Paltrow of the food world. They’d make the most stylish vegetable list, even wearing a pink ballgown three sizes too big. b. The mushroom-stuffed ravioli was the Gwyneth Paltrow of pastas: smooth, blond and lightweight, paired with white asparagus spears, poached quail eggs...


c. Heartbeat is simple, sophisticated and cool. It’s the Gwyneth Paltrow of handbags. d. Small Car. BMW Mini Cooper. The Gwyneth Paltrow of small cars: racy, stylish, and oh, so fashionable. (13) He insisted, however, that he was just the drummer, and Ulysses was the neurotic one. “I’m the Woody Allen of techno,” quipped Ulysses. As shown by these examples, the mention of the properties serving as utilized correspondences (Kövecses 2002: 79ff) in metaphorical mappings may do two kinds of jobs. Firstly, we may single out some central properties in the frame of the metaphorical source (our figuratively used proper name) which the bearer of these properties may possess to an extreme degree, as actually happens in the two examples in (10). The greater the conceptual distance between the two domains, the more useful the explicit mention of the utilized correspondences becomes, and the less we can rely on the paragon model. While the properties linked in the paragon model belong to the same frame, with metaphorical mappings between more distant domains, the properties linked by the utilized correspondences are not exactly the same. At best they are analogous, as in the Michael Jordan example, but may be quite different, as in the Shakespeare example. Secondly, the properties serving as utilized correspondences that are mentioned in the examples in (11–13) are certainly less central. They are additional, ancillary bits of information, often seen from a very subjective perspective, and they can be even ephemeral, e.g. something that has to do with only one episode in the life of the person whose name is used figuratively. For instance, Gwyneth Paltrow is of course not the only actress that can be characterized as smooth, blond and lightweight, but the speaker may perceive them as being sufficiently contrastive and relevant, and perhaps even outstanding, in the link established between the two domains. However, there is no implication in this group of examples that all or any of the properties in question are necessarily outstanding in the sense of being exhibited to the maximum degree. In fact, we may assume that it is quite likely that the properties in question will not be exhibited to the maximum, if there is need to explicitly specify them in the context, and that specifying such utilized correspondences will usually effect the cancellation of the correspondences compatible with the paragon model that we would have otherwise tried to establish. The axiological notions ‘best of ’ and ‘worst of ’ that we pointed out above as the most problematical in the paragon model arise in the final metonymic tier due to the imposition of a scalar model (Israel 1997, 1998) on these contrastive properties. The scalar model allows the metonymic mappings of the type whole scale for upper/lower end of scale (cf. Radden and Kövecses 1999: 32), whereby the property is interpreted as being exhibited to the maximum, either in the positive


Zidane


football

Figure 3. Scalar model imposed on contrastive properties and the metonymic mappings of the type whole scale for upper/lower end of scale

or negative sense. The scalar model and this type of metonymy have been shown to motivate a number of hyperbolic expressions in Brdar-Szabó and Brdar (2005). In Figure 3, the scalar model is indicated by the double arrow, while the cylindershaped figure represents the ICM containing the contrastive properties of Zidane the footballer, on which the scalar model is imposed. Typically, the utilized correspondences, if central in the frame, are left implicit. It appears that when the outcome of this final metonymic tier is used as input for metaphoric mappings, the more distant the two domains, the more likely it becomes that the axiological elements ‘best of ’ and ‘worst of ’ will become more dominant, occasionally so much so that the more objective standards of metaphorical comparison may become practically obliterated. Returning to the paragon-like outcome of the series of metonymic tiers ending with the whole scale for upper/lower end of scale metonymy, we observe that it is available for further elaboration, presumably metonymic mappings are involved in producing an instance of the metonymic type, as shown by the following examples: (14) a. It doesn’t take an Einstein to understand that this is a very important game for us… b. Soft-liners like Senator Joe Biden and consterned scientists will grump, but their ranks are thinned by the growing prospect of a Saddam or independent terrorist buying or building a nuke that could take out a U.S. city. c. Like all Bush women, she winks that she’s pro-choice. She explains how the president is actually for stem-cell research despite not being for it. She describes a Gary Cooper, not a John Wayne, pacing the South Lawn, agonizing over his decision to go to war.


d. In this short comedy by Cambridge’s Julian Joslin, poor Max is a miniWoody Allen, trudging lovelorn through the Harvard Square snow, sneaking peeks at Cosmo Girl! magazine,… The series of metonymic mappings summarized in Figure 3 can eventually form the input for metaphorical mappings, as illustrated in examples in (6), (10b), and (11– 13). In (6a) and (6b) we have metaphorical mappings between two conceptually relatively distant domains, although they are somewhat unusual because they proceed from the domain of humans to the domain of animals. Examples in (10b) and (11), (12) and (13) are similar in that respect. However, in (10a) it is not immediately clear that we have a metaphor: two domains are involved that are fairly closely related, we might even say, conceptually parallel and equally abstract or concrete. The question is whether this is also a case of metaphor, since this would be a sportto-another-sport type of mapping, which is odd, because in metaphors we typically have two distinct domains, both in terms of domain sort and domain concreteness or abstractness. Such examples abound in the domain of sports: (15) Gasquet will be the Zidane of French tennis. We would like to claim that these should also count as metaphors because we have distinct domains brought into correlation. The domains of basketball and rugby in (10a), and of French football and of French tennis in (15), are brought into correlation by another round of metonymic mapping taking as its input the result of yet another tier of metonymic mappings. In this tier, excellent qualities with respect to a given sport may be generalized (i.e. quite depleted in terms of meaning) by means of the high-level conceptual metonymy specific for generic (discussed in Radden and Kövecses 1999: 34). The target of this metonymic tier is something like ‘excellent qualities in general,’ and consequently the proper name comes to refer to a type of person rather than an actual person. The distance between the two domains is even smaller in the following example: (16) Micoud is the Zidane of the Bundesliga. If names of countries in the following examples are also taken as metonymies of a kind, viz. Spain and Israel seen as standing for Spanish football and Israeli corporate business life, respectively, then they are comparable to (10a): (17) a. The ‘Zidane of Spain’, as some dub Valeron, was perhaps destined to be a ‘super substitute’, brought on only when in desperate need of… b. Gil Shwed—the Bill Gates of Israel—made his fortune trading on corporate paranoia. Figuratively used proper names can also be applied within the primary domains of reference with postmodifiers making reference to time, as in (18):


(18) a. His keepy-uppies have become one of Scottish footballs greatest memories. The Zidane of the 60s, he has the ability to be the best player in this team. b. I call him “Little Bill” because I think this kid is maybe going to be the Bill Gates of this millennium. c. Edison was the Bill Gates of his day. Assuming that time is not a distinct domain in such cases but a dimension inherent in encyclopedic domains of Zidane and Bill Gates, respectively, we can easily imagine a different type of knowledge base resolution, i.e. along the time axis, which can be projected beyond the present time as well, i.e. into the as yet virtual section of the future. Although similar to examples (8–9), these clearly involve a metaphoric tier as well, which is then again followed by a metonymic conceptual partition of the sort we witness in (8–9). Regardless of whether we consider constructions in (15–18) as involving a metaphor topping a number of metonymic tiers or as involving another metonymic tier, all these expressions are again available for further elaboration, paralleling the examples in (14), the outcome now being an indefinite instance of the metonymic/ metaphtonymic type, as shown by the following examples: (19) a. It will take a Bill Gates of cars to solve our energy and transportation issues. b. Today, Naismith would be universally recognized as a genius, a Bill Gates of sport. c. Sonny is an Andy Warhol or Ted Turner or perhaps a Bill Gates of the literary world. d. the figure of Thomas Krens, the aggressive director of the Guggenheim Museum, who became somewhat of a Donald Trump or a Bill Gates of the museum world.

4. Conclusions and prospects for further research In the present chapter we have attempted to shed some light on the process of the construction of figurative meanings of proper nouns denoting humans. We have indicated why objectivist philosophical theories of meaning are doomed to fail in such an enterprise. By assuming that their meanings are not rigidly fixed in advance and outside the context of language use, and that they can be worked out only by enlisting our total encyclopedic knowledge organized in complex matrixes of domains, we have been able to show how it is possible to arrive at the various shades of meanings found with figuratively used proper nouns.


The process of such online construction of figurative meaning has been shown to be complex, dynamic and flexible, and its output can be revised at every step, further enriched with information, or subsequently depleted of it, depending on the cognitive operations employed. We have outlined how figurative meanings arise in a step by step fashion, involving tiers of metonymic mappings interspersed with metaphoric mappings. In doing so we have also demonstrated that the paragon model can be elaborated in such a way that specific mappings become well motivated.

References Anderson, John Mathieson. 2003. On the structure of names. Folia Linguistica 37: 347–98. Anderson, John Mathieson. 2004. On the grammatical status of names. Language 80(3): 435– 74. Barcelona, Antonio. 2003. Names: A metonymic ‘return ticket’ in five languages. Jezikoslovlje 4(1): 11–41. Barcelona, Antonio. 2004. Metonymy behind grammar: The motivation of the seemingly ‘irregular’ grammatical behavior of English paragon names. In Studies in Linguistic Motivation, Günter Radden and Klaus-Uwe Panther (eds), 357–74. Berlin: Mouton de Gruyter. Belaj, Branimir. 2005. On the cognitive status of mental spaces and of some types of metonymy within Conceptual Integration Theory. Jezikoslovlje 6(2): 119–44. Brdar, Mario and Brdar-Szabó, Rita. 2001. Vlastita imena između metonimijske Scile i meta foričke Haribde. Rasprave Instituta za hrvatski jezik i jezikoslovlje 27: 31–48. Brdar-Szabó, Rita and Brdar, Mario. 2005. Scalar models in a cognitive approach to hyperbolic expressions: With a little help from metonymy. In Pragmatics Today [Łódź Studies in Language 12], Piotr Cap (ed.), 75–94. Frankfurt: Peter Lang. Buljan, Gabrijela and Gradečak-Erdeljić, Tanja. 2003. The role of blending in some Croatian and English idiomatic structures. Paper read at the 8th International Cognitive Linguistics Conference, Logroño, La Rioja, July 20–5, 2003. Frege, Gottlob. 1952. On sense and reference. In Translations of the Philosophical Writings of Gottlob Frege, Peter Geach and Max Black (eds), 56–78. Oxford: Blackwell. Goossens, Louis. 1990. Metaphtonymy: the interaction of metaphor and metonymy in expressions for linguistic action. Cognitive Linguistics 1: 323–40. Israel, Michael. 1997. The scalar model of polarity sensitivity. In Negation and Polarity: Syntax and Semantics, Danielle Forget et al. (eds), 209–29. Amsterdam: John Benjamins. Israel, Michael. 1998. The Rhetoric of Grammar: Scalar Reasoning and Polarity Sensitivity. PhD dissertation, UCSD. Kleiber, Georges. 2004. Peut-on sauver un sens de denomination pour les noms propres? Functions of Language 11(1): 115–45. Kövecses, Zoltán. 2002. Metaphor: A Practical Introduction. Oxford: Oxford University Press. Kövecses, Zoltán and Radden, Günter. 1998. Metonymy: Developing a cognitive linguistic view. Cognitive Linguistics 9(1): 37–77. Kripke, Saul. 1972/1980. Naming and Necessity. Cambridge MA: Harvard University Press. Lakoff, George. 1987. Women, Fire, and Dangerous Things: What Categories Reveal about the Mind. Chicago IL: University of Chicago Press.

142 Mario Brdar and Rita Brdar-Szabó Lakoff, George and Johnson, Mark. 1999. Philosophy in the Flesh: The Embodied Mind and Its Challenge to Western Thought. New York NY: Basic Books. Lehrer, Adrienne. 1999. Proper names: Linguistic aspects. In Concise Encylopedia of Grammatical Categories, Keith Brown and John Miller (eds), 311–13. Amsterdam: Elsevier. Mill, John Stuart. 1843. A System of Logic, Ratiocinative and Inductive: Being a Connected View of the Principles of Evidence, and Methods of Scientific Investigations. London: Parker. Panther, Klaus-Uwe. 2005. Metonymic reasoning inside and outside language. LACUS Forum XXXI: Interconnections. Adam Makkai, William J. Sullivan, and Arle R. Lommel (eds), 15– 32. Houston TX: The Linguistic Association of Canada and the United States. Panther, Klaus-Uwe and Thornburg, Linda. 2004. The role of conceptual metonymy in meaning construction. metaphorik.de 6: 91–116. Paradis, Carita. 2004. Where does metonymy stop? Senses, facets, and active zones? Metaphor and Symbol 19(4): 245–64. Radden, Günter and Kövecses, Zoltán. 1999. Towards a theory of metonymy. In Metonymy in Language and Thought, Klaus-Uwe Panther and Günter Radden (eds), 17–5. Amsterdam: John Benjamins. Russell, Bertrand. 1956. The philosophy of logical atomism. In Logic and Knowledge, Robert Marsh, (ed.), 177–281. London: George Allen and Unwin. Thurmair, Maria. 2002. Der Harald Juhnke der Sprachwissenschaft: Metaphorische Eigennamenverwendungen. Deutsche Sprache 30(1): 1–27.

chapter 7

Collocational overlap can guide metaphor interpretation* Anatol Stefanowitsch University of Bremen

In this chapter, I argue that information about properties that are metaphorically projected between semantic domains is accessible in the form of collocational overlap between lexical items representing these domains. This shows that the statistical properties of natural language can guide human language users (and potentially any language-processing device) in constructing metaphorical meaning. Keywords: quantitative corpus linguistics, statistical language processing, metaphor, collocational overlap, meaning construction

1. Introduction It is by now widely accepted that metaphor is fundamental to natural language and that the interpretation of metaphorical expressions constitutes a core aspect of meaning construction. In this chapter, I argue that some of the information necessary to construct metaphorical meanings is present in the collocational overlap between literally and metaphorically used lexical items cooccurring in metaphorical expressions. This information can guide the human language user as well as non-human language processing devices in assigning interpretations to metaphorical statements. Linguistic theories of metaphor typically derive the motivation for individual metaphorical expressions from large-scale connections between entire semantic domains. Generally, these theories assume projections from one semantic domain (variously referred to as source domain, image donor, secondary subject, etc.) onto another (the target domain, image recipient, primary subject, etc.) (see, for example, * I would like to thank, first of all, Klaus-Uwe Panther, who, for obvious reasons, never saw this paper before it was published but without whose encouragement and guidance I would not be a linguist, let alone a cognitively oriented linguist, today. I would also like to thank the editors of this volume, especially Thomas Berg, for their valuable input and their gentle but unyielding insistence that I include an elicitation experiment, thus making the paper more “cognitive,” and, hopefully, more substantial. I alone am responsible for any conceptual or methodological shortcomings of my implementation of their suggestions or any other aspect of this chapter.

144 Anatol Stefanowitsch

Black 1962, 1993[1979]; Weinrich 1976; Lakoff and Johnson 1980; Lakoff 1993). Such projections manifest themselves most obviously in equative statements such as the following, which may thus serve to illustrate some basic aspects of the interpretation of metaphorical language (or, to put it differently, of the construction of metaphorical meaning):

(1) a. Man is a wolf (Black 1993[1979]: 20) b. Richard is a gorilla (Searle1993 [1979]: 92) c. Sam is a pig (Searle 1993[1979]: 83)

The interpretation of such statements requires at least two kinds of (declarative) knowledge from the language user, one of which this chapter will discuss in detail. First, the language user must have some sort of knowledge of the metaphorical mappings that exist in the language in question. For example, the interpretation of (1a–c) involves, among others, the mapping human behavior is animal behavior. The extent and nature of this metaphorical knowledge has been extensively (but far from conclusively) addressed in the cognitive linguistic literature over the past twenty-five years and continues to be a major concern to this day. I will have little to say about this issue here (but cf. Stefanowitsch 2005 for an initial corpusbased exploration). Second, and crucially in the context of this chapter, the language user must know which aspect(s) of the source domain (e.g. wolf, pig, etc.) are projected onto the target domain (e.g. human beings; in Ungerer and Schmid’s (2006) terms, they must have information about the mapping scope). Clearly, not every aspect can be (or is likely to be) projected. For example, the interpretation of (1a) involves the knowledge that it is the (supposedly) evil or dangerous nature of wolves that is projected onto human beings, rather than their social behavior, their mating habits or the fact that they are an endangered species. This issue has not been dealt with anywhere near as extensively in the literature as has the nature of metaphorical knowledge (but see further below). In addition, the language user must have procedural knowledge as to how the declarative knowledge is to be applied in the actual process of on-line meaning construction. There seem to be two major theories concerning this issue, one that gives primacy to pragmatic inferencing (e.g. Searle 1993[1979]) and one that gives primacy to the activation of cognitive models (e.g. Gibbs 1994; cf. also Panther and Thornburg 1998 on metonymy). As far as I can tell, the two approaches are not mutually exclusive, but I will not pursue this issue. The suggestions I develop below are certainly compatible with both views. In this chapter, I will suggest a collocational overlap model of metaphor interpretation. As its name suggests, the overlap model is based on the idea that the process of metaphor interpretation can be guided by knowledge concerning the colloca-

Collocational overlap can guide metaphor interpretation 145

tional overlap displayed by central lexical items from the source and target domains. By collocational overlap of two lexical items I mean the set of (significant) collocates shared by these items (cf. e.g. Gries 2004; Wulff 2006), rank-ordered in such a way that the collocates most important to both items score highest. Specifically, I assume that any language processor (whether human or machine) can deduce from prior exposure to the language the probability with which a given collocate occurs with a given node word. Upon hearing an utterance linking lexical items from two domains such that one item is predicated of the other, the processor activates the collocates of both lexical items and computes for each collocate its combined probability of occurrence with both items. Such an utterance may be an equative statement such as those presented in (1a–c), but it may also take a range of other forms (cf. below). Those collocates with the highest combined probability then offer crucial clues as to which aspects of the source domain are likely to be transferred to the target domain. Let me stress two points, so as not to be misinterpreted concerning the scope of this model. First, I do not claim that metaphor interpretation can be exclusively accounted for in terms of collocational overlap. If psychologically oriented theories of metaphor are right, as I assume they are, metaphorical mappings transfer entire networks of world knowledge from one domain to another, rather than just individual words. However, I argue that this transfer can be guided by the identification of collocational overlap such that this overlap provides a clue as to which aspects of the source domain are most relevant to the mapping as a whole. Second, I do not claim that it is only the properties referred to by the shared (or most strongly shared) collocates that are transferred from the source to the target domain. If this were the case, there would be not much point in using metaphorical expressions at all. However, I argue that the shared properties may constitute the conceptual core that makes the mapping possible in the first place. I will present three variants of the collocational overlap model: (i) a symmetrical variant that gives equal weight to the source and the target domain, (ii) a ‘sourcedominant’ asymmetrical variant that gives more weight to the source-domain collocates, and (iii) a ‘target-dominant’ asymmetrical variant that gives more weight to the target-domain collocates. The symmetrical variant is intended to demonstrate the basic plausibility of the overlap model, making a minimum of additional assumptions. The two asymmetrical variants are intended to correspond very roughly to two proposals from the literature concerning the way in which properties relevant for a mapping are distinguished from irrelevant ones. The source-dominant version reflects the idea that the source domain is in its entirety mapped onto the target domain, but that there are ‘target-domain overrides’ that filter out those properties of the source domain that are simply not applicable in the target domain (cf. Lakoff 1993). The target-dominant version reflects the idea that metaphorical map-


pings are interactive processes by which relevant target-domain properties are first selectively projected onto the source domain, and then transferred back to the target domain (cf. Black 1962, 1979). Clearly, the various versions of the overlap model do not capture the theories proposed by Lakoff and Black in their full complexity; they are simply different implementations of the overlap model in light of these theories. Before I turn to the collocational overlap model in more detail, let me briefly return to the issue of the form of metaphorical utterances hinted at above. Clearly, metaphorical utterances take all kinds of forms, and equative statements like those in (1a–c) may not even be among the most frequent of these forms. Some other formal realizations of the metaphorical mappings manifest in the expressions in (1a– c) are shown in (2):

(2) a. b. c.

He grinned wolfishly (TD as subject, SD as adverb modifying the predicate) He has gorilla arms (TD as subject, SD as noun modifying the predicate noun) He pigged out on ice cream (TD as subject, SD as predicate)

The model developed below applies to the interpretation of metaphorical expressions of all kinds, those in (1a–c) just as well as those in (2a–c). The expressions in (1a–c) are simply used as a shorthand way of referring to all metaphorical expressions that explicitly combine source and target-domain vocabulary (such expressions are referred to as metaphorical patterns in Stefanowitsch 2004, 2005, 2006). The chapter is structured as follows. In Section 2, I present a simple, frequencybased implementation of the overlap model and apply it to the statements in (1a–c). In Section 3, I present a slightly more sophisticated implementation based not on the simple frequency of collocates but on their association strength relative to the node. Finally, in Section 4, I present a simple elicitation study aimed, first, at assessing the quality of the models in terms of the way in which speakers actually interpret statements such as (1a–c), and second, at determining whether the overlap model can predict the interpretation assigned to non-conventionalized metaphorical expressions, such as Sam is a dolphin.

2. A frequency-based model of collocational overlap 2.1 Developing the model Aims. In this section, I will introduce a simple method for calculating the collocational overlap between two node words, more precisely, for identifying the shared collocates and the degree to which they are relevant within the overlap.


I will assume a simple, frequency-based definition of collocation: a word is considered to be a collocate of the node word to the degree that it occurs frequently in the context of this node word. Such a definition is known to be problematic (see Section 3), but it is mathematically simple and does not require any theoretical assumptions beyond the notion that there are frequency effects in language. Data and method. The most obvious choice of word class for the collocates denoting properties of a noun is, of course, the class of adjectives. There are two potentially promising grammatical contexts in which to look for adjectives: preceding the head noun (attributive position, as in bad wolf, endangered wolves) or following a copula with the head noun as subject (predicative position, as in the wolf is bad, wolves are endangered). Theoretically, it should not make any difference which of these positions is chosen, but in practice there are at least two reasons for choosing the latter. First, we can expect the number of relevant hits in an untagged corpus (see below) to be much higher for search strings like wolf is or wolves are than for search strings like wolf or wolves. For example, the British National Corpus (BNC) contains 778 hits for wolf/wolves, only 121 (i.e. 15.5%) of which are preceded by an adjective, and 8 hits for wolf is/wolves are, 3 (i.e. 37.5%) of which are followed by an adjective). Second, strings instantiating patterns like [Adj + wolf] are relatively likely to be metaphorical uses, as in [M]ultinational corporations ... are big bad wolves (more than a quarter of all hits in the BNC are of this type), whereas strings instantiating [wolf BE Adj] are unlikely to be metaphorical uses (all hits in the BNC are literal). Clearly, metaphorical uses would contaminate a fair assessment of the properties associated with real wolves. Thus, although the problems with attributive uses are clearly not insurmountable, I will limit myself to the predicative context for the present study. I further decided to limit the context to plural nouns and to statements using the simple present tense, in both cases in order to maximize the number of hits with generic reference. Choosing a suitable corpus proved to be the most problematic aspect of the research design. As will have become clear, the largest balanced corpus currently available, the 100-million-word BNC, contains far too few occurrences in the relevant context (for example, only 5 hits for wolves are). The only larger corpora currently available are newspaper corpora such as the North American News corpus and the Reuters corpus, which also do not contain enough occurrences, presumably due to the limited range of text types they include. I therefore decided to use the Internet as a corpus. In order to maximize the number of hits produced by native speakers of English, only websites with the country suffixes (for Great Britain) and (for the United States of America) were used. The websites were accessed via the search engine Google. The search patterns followed the format [“wolves are” site:uk OR site:us] and the results were then manually post-edited to extract only those hits where the search string was followed by an adjective modifying the noun in ques-

148 Anatol Stefanowitsch Table 1. Frequency list of adjectives occurring in the pattern [men are ADJ] Rank

Adjective

passoc

Rank

Adjective

passoc

Rank

Adjective

passoc

1. 2. 3. 4. 5. 6. 7.

happy aware reluctant able involved prone dangerous

0.0279 0.0279 0.0244 0.0209 0.0209 0.0209 0.0174

8. 9. 10. 11. 12. 13. 14.

old free strong good at different equal bad

0.0174 0.0174 0.0100 0.0139 0.0139 0.0105 0.0105

15. 16. 17. 18. 19.

concerned tall good violent crap

0.0105 0.0105 0.0105 0.0105 0.0105

Table 2. Frequency list of adjectives occurring in the pattern [wolves are ADJ] Rank

Adjective

passoc

Rank

Adjective

passoc

Rank

Adjective

passoc

1. 2. 3. 4. 5. 6. 7. 8.

social large intelligent endangered hungry dangerous extinct shy

0.0426 0.0398 0.0313 0.0313 0.0284 0.0284 0.0256 0.0227 .

9. 10. 11. 12. 13. 14. 15. 16.

afraid strong interesting wild bad protected beautiful good

0.0199 0.0199 0.0199 0.0170 0.0170 0.0170 0.0170 0.0142 .

17. 18. 19. 20. 21. 22. 23. 24.

cool evil mean small common gray big smart

0.0114 0.0114 0.0114 0.0114 0.0114 0.0114 0.0114 0.0114

tion. An adjective was considered to fulfill this requirement (i) if it headed a subject complement AP, as in wolves are endangered, wolves are very intelligent and social; or (ii) if it modified the head noun of a subject complement NP where the head noun constitutes a superordinate term of the subject NP, such as wolves are very intelligent animals or wolves are an endangered species (animal and species are superordinates of wolf ) but not wolves are good hunters (hunter is not a superordinate of wolf ). Tables 1 and 2 list the most frequent adjectives occurring in the pattern [men are ADJ] and [wolves are ADJ] respectively, where passoc is simply the likelihood that (or relative frequency with which) an adjective occurs in the pattern in question. Note, first of all, that the adjectives in Table 2 show quite clearly that a definition of metaphorical mappings as all-encompassing one-way projections from a source to a target domain would lead to the wrong predictions concerning the meaning of Man is a wolf. If metaphorical mappings were such one-way projections, a simple source-domain frequency list like that in Table 1 should have adjectives denoting the mapped properties in top positions. In other words, if all properties associated with wolf were simply projected onto man in order to interpret this utterance, then we would expect it to mean ‘Man is social, large, intelligent, and/or endangered, etc.’ However, the properties actually ascribed to men via this utterance are those referred to by adjectives like fierce, predatory, rapacious, evil, dangerous, and their near synonyms (these are the adjectives used in defining the metaphorical meaning


of wolf in five major dictionaries of British or American English, cf. LDE, CALD, NHD, MWO, AHD, s.v. wolf ). Notice that these adjectives and their synonyms do in fact occur on the sourcedomain frequency list for [wolves are ADJ] (e.g. dangerous, bad, evil, mean, and perhaps also large, wild, and big; cf. Table 2) and on the target-domain frequency list for [men are ADJ] (e.g. dangerous, bad, violent, and perhaps strong; Table 1). However, they are not among the most frequent adjectival collocates for either of the two words. If the overlap model is right, however, they should be among the highestranking overlapping collocates, i.e. the collocates shared by men and wolves. The simplest way of modeling the way in which a language processor might identify the overlap between the collocates of the two words is by combining the individual probabilities, as shown in (3):

(3) Symmetrical overlap pcombined =P (adj|man) × P (adj|wolf )

This formula is a straightforward translation of the collocational overlap model described in the introduction: upon hearing an utterance like man is a wolf (or the man grinned wolfishly, etc.), the processor simply computes for each adjective the combined probability of occurrence with both nouns. Those adjectives with the highest combined probabilities are the most likely candidates for encoding the relevant properties. Note that this formula also ensures that only those collocates that occur with both node words are included in the overlap; all other collocates will automatically have a combined probability of zero. This symmetrical version of the overlap model is overly simple in one fundamental respect. Metaphorical mappings and the expressions instantiating them are not semantically symmetrical: man is a wolf does not mean the same thing as wolf is a man. The formula in (3) simply does not allow us to distinguish between these two statements. We can take this asymmetry into account, however, by weighting one set of probabilities more heavily than the other. As mentioned, Black’s interaction theory seems to attribute a more important role to the target domain, with the source domain acting more or less like a filter: properties that are highly relevant to the target domain are projected onto the source domain and then back. Conversely, Lakoff ’s theory of target-domain overrides seems to attribute a more important role to the source domain: properties associated with the source domain are projected onto the target domain, which filters out only those properties that do not apply in the target domain. Since we are dealing with probabilities (i.e. decimal fractions), a relatively straightforward way of increasing the weight of one of the probabilities is to take its square root (e.g. √0.9 = 0.9487). Unlike simpler mechanisms conceivable, such as multiplication by 2, using the square root will ensure that the probability stays in


Table 3. Frequency-based overlap between [men are ADJ] and [wolves are ADJ] Symmetrical Rank 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31.

Adjective dangerous strong intelligent bad good free smart afraid numerous old sensitive good at common tall powerful scary similar noble aggressive dead active useless stupid close nice quiet satisfied brave restless present wise

Source-Dominant pcomnbined 4.95E–04 2.77E–04 2.18E–04 1.78E–04 1.48E–04 9.90E–05 7.92E–05 6.93E–05 5.94E–05 4.95E–05 3.96E–05 3.96E–05 3.96E–05 2.97E–05 2.97E–05 2.97E–05 2.97E–05 1.98E–05 1.98E–05 1.98E–05 1.98E–05 1.98E–05 1.98E–05 1.98E–05 9.90E–06 9.90E–06 9.90E–06 9.90E–06 9.90E–06 9.90E–06 9.90E–06

Rank 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31.

Adjective dangerous intelligent strong bad good afraid smart free numerous common powerful similar scary sensitive old active good at aggressive noble tall stupid useless close dead nice restless present wise quiet satisfied brave

Target-Dominant pcomnbined 3.75E–03 2.61E–03 2.35E–03 1.74E–03 1.45E–03 1.17E–03 9.49E–04 7.50E–04 7.11E–04 6.71E–04 5.03E–04 5.03E–04 5.03E–04 4.74E–04 3.75E–04 3.35E–04 3.35E–04 3.35E–04 3.35E–04 2.90E–04 2.37E–04 2.37E–04 2.37E–04 2.37E–04 1.68E–04 1.68E–04 1.68E–04 1.68E–04 1.68E–04 1.68E–04 1.68E–04

Rank 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31.

Adjective dangerous strong bad free good intelligent old good at smart numerous tall sensitive afraid stupid common useless dead close similar scary powerful noble aggressive active nice quiet satisfied brave restless present wise

pcomnbined 2.94E–03 1.97E–03 1.36E–03 1.31E–03 1.25E–03 1.23E–03 9.29E–04 7.43E–04 7.43E–04 6.43E–04 5.57E–04 5.25E–04 4.91E–04 3.71E–04 3.71E–04 3.71E–04 3.71E–04 3.71E–04 3.22E–04 3.22E–04 3.22E–04 2.63E–04 2.63E–04 2.63E–04 1.86E–04 1.86E–04 1.86E–04 1.86E–04 1.86E–04 1.86E–04 1.86E–04

the range between 0 and 1. This strategy will give us the following two formulas:

(4) Source-domain-dominant overlap (‘Target domain overrides’) pcombined = P (adj|man) × √P (adj|wolf )

(5) Target-domain dominant overlap (‘Interaction theory’) pcombined = √P (adj|man) × P (adj|wolf )

. Of course, taking the square root increases the weight of smaller fractions more strongly than that of larger fractions; it remains to be seen whether this is a problem.

Collocational overlap can guide metaphor interpretation

It must be emphasized once again that I am not attempting to reduce Black’s and Lakoff ’s theories to the model proposed here or the formulas in (4) and (5). It is obvious that there is more to these theories than is captured by the overlap model, let alone by the formulas given here. Instead, the formulas should simply be understood as one of several ways of operationalizing the theories from a corpus-based perspective. Results. The results of the procedure described above are shown in Table 3. For each of the three versions, the overlapping adjectival collocates are listed in decreasing order of combined probability. Discussion. All three versions of the method seem promising. The top five overlapping collocates now include adjectives highly relevant to the interpretation of man is a wolf: dangerous, strong, and bad. The adjective intelligent can at least be said not to contradict the intended interpretation: Man is a wolf does seem to refer to an intelligent kind of badness (in contrast to, e.g. that guy is a gorilla). Of course, the presence of the adjective good among the top five collocates spoils this picture to a certain extent, although it is ranked below bad by all three methods. Also, some relevant adjectives are very far down the list (for example, scary and aggressive). This is not a major problem, as one would certainly not expect all potentially relevant collocates to end up among the top five or ten, but clearly the larger the proportion of relevant adjectives among the top collocates, the more convincing the model will appear. If we want to rank the different versions of the model informally, the target-domain dominant version seems to fare best: it places three of the most clearly relevant adjectives in the top three positions (as opposed to two for the other two versions), and bad is ranked most clearly above good (two places above, as opposed to one place above in the other two models). Interestingly, it also ranks free among the top five collocates. This may at first glance seem problematic, but in fact there is a metaphorical use of wolf that is associated with freedom: He is a lone wolf.

2.2 Testing the model Aims. In order to test the generalizability of the model developed above, I will apply it to the other two source domain items in (1): gorilla and pig (cf. examples (1b–c) . I will hold the target domain (man) constant. Data and Method. The corpus again consisted of and websites accessed via the search engine Google. The search strings were gorillas are, pigs are, and ice is. The same rules as before were applied in order to extract the relevant adjectives, and the formulas in (3), (4), and (5) were applied. Results. The most frequent adjectives in the pattern [gorillas are ADJ] are shown in Table 4. The relevant adjectives, according to the dictionaries mentioned above, are large, violent, ugly, brutal, brutish, or their near synonyms. Intuitively, we might add stupid to this list. Almost no relevant adjectives occur among the top twen-

151


Table 4. Most frequent adjectives occurring in the pattern [gorillas are ADJ] Rank

Adjective

passoc

Rank

Adjective

passoc

Rank

Adjective

passoc

1. 2. 3. 4. 5. 6. 7.

endangered large big strong gentle threatened shy

0.0763 0.0678 0.0424 0.0381 0.0339 0.0297 0.0297

8. 9. 10. 11. 12. 13. 14.

peaceful sensitive intelligent social dangerous vegetarian stupid

0.0254 0.0254 0.0254 0.0212 0.0212 0.0169 0.0127

15. 16. 17. 18. 19. 20. 21.

smart dark interesting close black plant-eating protected

0.0127 0.0127 0.0127 0.0127 0.0127 0.0127 0.0127

ty adjectival collocates of gorillas are, which again shows that a simple projection model could not predict the right interpretation. Table 5 shows the overlapping collocates of [men are ADJ] and [gorillas are ADJ]. Again, the results are promising if we compare them to the frequency list in Table 4. The top ten overlapping collocates for each version of the model include the relevant adjectives strong, dangerous, and stupid, with the former two occupying the top two positions. However, the results are clearly less convincing than in the case of wolf: the top ten collocates also contain adjectives that are completely contrary to the right interpretation: sensitive, intelligent and smart. Intelligent and (in the case of the target-dominant method) smart are even ranked above stupid. Table 5. Frequency-based overlap between [men are ADJ] and [gorillas are ADJ] Symmetrical Rank 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21.

Adjective strong dangerous sensitive intelligent reluctant close stupid smart different susceptible capable active powerful aggressive quiet present protective intellectual scary insensitive nice

Source-dominant pcombined 5.32E–04 3.69E–04 1.77E–04 1.77E–04 1.03E–04 8.86E–05 8.86E–05 8.86E–05 5.91E–05 5.91E–05 5.91E–05 2.95E–05 2.95E–05 2.95E–05 2.95E–05 1.48E–05 1.48E–05 1.48E–05 1.48E–05 1.48E–05 1.48E–05

Rank 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21.

Adjective strong dangerous reluctant intelligent sensitive different stupid smart close susceptible capable aggressive active powerful quiet present protective scary insensitive intellectual nice

Target-dominant pcombined 2.72E–03 2.54E–03 1.59E–03 1.11E–03 1.11E–03 9.07E–04 7.86E–04 7.86E–04 7.86E–04 6.42E–04 6.42E–04 3.21E–04 3.21E–04 3.21E–04 3.21E–04 2.27E–04 2.27E–04 2.27E–04 2.27E–04 2.27E–04 2.27E–04

Rank 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21.

Adjective strong dangerous intelligent sensitive smart stupid close capable susceptible reluctant active quiet different aggressive powerful insensitive intellectual nice present protective scary

pcombined 4.50E–03 2.80E–03 2.12E–03 2.12E–03 1.06E–03 1.06E–03 1.06E–03 7.07E–04 7.07E–04 6.62E–04 5.00E–04 5.00E–04 5.00E–04 5.00E–04 5.00E–04 2.50E–04 2.50E–04 2.50E–04 2.50E–04 2.50E–04 2.50E–04


Table 6. Most frequent adjectives occurring in the pattern [pigs are ADJ] Rank

Adjective

passoc

Rank

Adjective

passoc

Rank

Adjective

passoc

1. 2. 3. 4. 5. 6. 7.

susceptible clean intelligent pink fat healthy dirty

0.0415 0.0415 0.0369 0.0276 0.0276 0.0230 0.0230

8. 9. 10. 11. 12. 13. 14.

able smart equal small aggressive present happy

0.0230 0.0184 0.0184 0.0138 0.0138 0.0138 0.0138

15. 16. 17. 18. 19.

common clever close popular cute

0.0138 0.0138 0.0138 0.0138 0.0138

Next, the most frequent adjectives in the pattern [pigs are ADJ] are shown in Table 6. The relevant properties, according to the dictionaries mentioned above, are fat, greedy, gross, impolite, offensive, unpleasant, difficult to deal with, unkind, untidy, and their synonyms. It must be noted that the metaphorical use of pig seems less consistent than in the case of wolf and gorilla: there are at least three semantic groups: excessive eating, antisocial behavior and messiness. Further evidence of this inconsistency comes from various specialized uses of pig to refer to immoral women, sexist or racist members of the establishment (sexist pig, racist pig), and (in some varieties) police officers. Finally, although none of the dictionaries mention this, the property happy may also be associated with pig (happy as a pig). This range of possible interpretations must, of course, be kept in mind. Three relevant adjectives occur among the top twenty adjectival collocates of [pigs are ADJ]: dirty, aggressive, and happy. These do not, however, occur in very high positions on the list, so that a simple projection model would again make incorrect predictions concerning the interpretation of men are pigs. Table 7 (p. 154) shows the overlapping collocates of [men are ADJ] and [pigs are ADJ]. Again, the results are promising compared to the simple frequency list but, as in the case of gorilla, they are not quite as convincing as in the case of wolf. The top ten overlapping collocates contain a number of relevant adjectives: happy, healthy, fat, and, in the case of the target-dominant method, dirty. Thus, the target-dominant method is once more slightly better than the other two. However, the top ten collocates also contain words that are contrary to, or at least irrelevant for, the correct interpretations. Concentrating on the target-domain dominant version, there are intelligent, smart, susceptible, able, equal, and close. What is especially problematic is that the top two collocates are irrelevant to the correct interpretation and that the most relevant adjectives do not occur among the top five collocates. In addition, some of the clearly relevant adjectives occur in relatively low positions (difficult, lazy, messy, and heavy). Discussion. In general, the results provide support for the collocational overlap model as compared to a simple source-domain frequency model. In all cases, adjec-

153


Table 7. Frequency-based overlap between [men are ADJ] and [pigs are ADJ] Symmetrical Rank 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37.

Adjective able happy susceptible intelligent equal healthy smart aware fat close dirty free dangerous different aggressive bad common present clever active lazy cold sensitive same available difficult capable funny heavy lean lost messy nice similar responsive noble young

Source-dominant pcombined 4.82E–04 3.85E–04 2.89E–04 2.57E–04 1.93E–04 1.61E–04 1.28E–04 1.28E–04 9.63E–05 9.63E–05 8.03E–05 8.03E–05 8.03E–05 6.42E–05 4.82E–05 4.82E–05 4.82E–05 4.82E–05 4.82E–05 3.21E–05 3.21E–05 3.21E–05 3.21E–05 3.21E–05 3.21E–05 3.21E–05 3.21E–05 1.61E–05 1.61E–05 1.61E–05 1.61E–05 1.61E–05 1.61E–05 1.61E–05 1.61E–05 1.61E–05 1.61E–05

Rank 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37.

Adjective happy able aware susceptible equal intelligent free dangerous healthy different smart close bad fat dirty sensitive cold difficult lazy capable common aggressive present clever available active same heavy lost funny lean young responsive messy similar noble nice

Target-dominant pcombined 3.28E–03 3.17E–03 1.89E–03 1.42E–03 1.42E–03 1.34E–03 1.18E–03 1.18E–03 1.06E–03 9.46E–04 9.46E–04 8.19E–04 7.10E–04 5.79E–04 5.29E–04 4.73E–04 4.73E–04 4.73E–04 4.73E–04 4.73E–04 4.10E–04 4.10E–04 4.10E–04 4.10E–04 3.35E–04 3.35E–04 3.35E–04 2.37E–04 2.37E–04 2.37E–04 2.37E–04 2.37E–04 2.37E–04 2.37E–04 2.37E–04 2.37E–04 2.37E–04

Rank 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36. 37.

Adjective susceptible able intelligent happy healthy equal fat smart dirty close clever common present aggressive aware free dangerous same different active available bad cold sensitive capable difficult lazy similar messy nice noble responsive funny lost young lean heavy

pcombined 3.46E–03 3.33E–03 3.08E–03 2.31E–03 1.92E–03 1.88E–03 1.63E–03 1.54E–03 1.36E–03 1.15E–03 8.16E–04 8.16E–04 8.16E–04 8.16E–04 7.69E–04 6.08E–04 6.08E–04 5.44E–04 5.44E–04 5.44E–04 5.44E–04 4.71E–04 3.85E–04 3.85E–04 3.85E–04 3.85E–04 3.85E–04 2.72E–04 2.72E–04 2.72E–04 2.72E–04 2.72E–04 2.72E–04 2.72E–04 2.72E–04 2.72E–04 2.72E–04

tives referring to the properties relevant to the interpretation feature prominently among the overlapping collocates, but not among the collocates of the source-domain expression by itself. In all four case studies, the target-domain dominant version of the model fared best, which may be interpreted as support for Black’s interaction theory.


Clearly, however, the picture is far from perfect. In all cases, there are individual adjectives among the top collocates which are either irrelevant to the conventional interpretation, or even contradict it (good for wolf, intelligent/smart for gorilla and pig). There are several possible reasons for this. First, the overlap model may simply be wrong. Of course, this possibility always exists and can certainly not be excluded on the basis of a single study. However, this possibility does not immediately suggest itself in light of the generally encouraging nature of the results. Second, the data may consist of text types that are not representative of the way we usually talk about the source domains in question. Given the fact that an unbalanced and uncontrollable corpus (the Internet) was used, this is certainly a possibility which must be taken seriously, but which is, unfortunately, beyond the scope of the present study. Finally, the simple frequency-based definition of collocation and of collocational overlap may not be a realistic operationalization of the overlap model. I will investigate this possibility in the next section.

3. An association-based model of collocational overlap 3.1 Developing the model Aims. In this section, I will refine the overlap model by using a more sophisticated definition of collocation, based on association strength rather than on frequency. The problem with a frequency-based definition is that it does not take into account the base frequency of the linguistic items in question and thus does not allow us to assess whether an item occurs more or less frequently with the node word than would be expected by chance alone (see e.g. Church and Hanks 1990, see also Stefanowitsch and Gries 2003). This means that collocates that are fairly highly ranked in terms of frequency may actually occur in the context of the node word simply because they are frequent in general, not because they are particularly strongly associated with the node word as such. For example, good is the sixteenth most frequent adjective in the frame wolves are ADJ, and in the collocational overlap with men are ADJ it ends up in fifth position, although it is contrary to the correct metaphorical interpretation of wolf. When we compare its observed frequency in wolves are ADJ with its expected frequency, however, we find that good occurs significantly less frequently in this frame than its base frequency would lead us to expect. Thus, an analysis that takes into account base frequencies would have identified good as an ‘anticollocate’ of wolf, and thus removed it from the list. Data and Method. The data were the same as before. The association-based probabilities were derived on the basis of the binomial test, which measures the likelihood that the deviation of an observed frequency of occurrence from its expected frequency is due to chance. In particular, I used the formula in (6), where b is the

155


cumulative binomial p-value for the occurrence of a given word in the critical context given its overall probability of occurrence, and where the sign of the logarithm is set to plus if the word occurs more frequently than expected and to minus if it occurs less frequently than expected: (6) Association-based probability ±log bi passoc = n ±log bi

Σ j=1

Taking the logarithm of the binomial p-value turns an exponential scale into a linear scale and dividing the value of a given adjective by the sum of the values of all adjectives in the critical context ensures that the values for passoc fall between 0 and 1 and that their sum is 1. The binomial p-value is used here because the overall probabilities of occurrence for the words could not be derived from the same data set as their probabilities of occurrence in the critical context (there is no straightforward way of determining the frequency of all adjectives on / websites). Instead, they were derived from Leech et al.’s (2001) lemmatized frequency list of the British National Corpus and then treated as a priori probabilities. In theory, of course, any p-value of an exact test (e.g. the Fisher-Yates exact test) could be used in the formula. Using frequencies from one corpus in order to derive base-line expectations about another corpus is of course not without problems, but since the BNC is a large and diverse corpus, expected frequencies derived from it should not be entirely off the mark. Tables 8 and 9 show the most strongly associated collocates for [men are ADJ], and [wolves are ADJ] respectively. Clearly, the rankings are similar to the frequency-based rankings in some details and different in others; for our purposes, what is most important is that, again, the collocates of wolf by itself cannot correctly predict its metaphorical interpretation. Table 8. Strongest adjectival collocates in the pattern [men are ADJ] Rank Adjective

passoc

Rank Adjective

passoc

Rank Adjective

passoc

1. 2. 3. 4. 5. 6. 7. 8. 9. 10.

3.34E–02 3.22E-02 3.10E-02 2.17E-02 2.00E-02 2.00E-02 1.55E-02 1.50E-02 1.44E–02 1.44E–02

11. 12. 13. 14. 15. 16. 17. 18. 19. 20.

1.44E–02 1.17E-02 1.10E-02 1.10E-02 1.10E-02 1.10E-02 1.10E-02 1.03E-02 1.00E-02 1.00E-02

21. 22. 23. 24. 25. 26. 27. 28.

9.60E–03 9.30E–03 9.30E–03 9.00E–03 8.50E–03 8.10E–03 8.10E–03 8.00E–03

prone reluctant good at aware crap happy involved dangerous good in reticent

promiscuous macho mortal sexist vain overweight disposable violent damaged affected

susceptible lazy rotten terrified ashamed tall useless free


Table 9. Strongest adjectival collocates in the pattern [wolves are ADJ] Rank Adjective

passoc

Rank Adjective

passoc

Rank Adjective

passoc

1. 2. 3. 4. 5. 6. 7. 8. 9. 10.

5.82E-02 4.61E-02 4.18E-02 3.72E-02 3.25E-02 2.67E-02 2.65E-02 2.34E–02 1.81E-02 1.70E-02

11. 12. 13. 14. 15. 16. 17. 18. 19. 20.

1.65E-02 1.50E-02 1.50E-02 1.49E-02 1.40E-02 1.35E-02 1.30E-02 1.28E-02 1.18E-02 1.15E-02

21. 22. 23. 24. 25. 26. 27. 28. 29.

1.10E-02 1.08E-02 1.08E-02 1.08E-02 1.07E-02 1.00E-02 9.10E–03 8.20E–03 8.20E–03

endangered extinct intelligent hungry shy dangerous protected gray social misunderstood

afraid scary adaptable large wild interesting evil smart abundant beautiful

mean socialised carnivorous sneaky vicious cool strong treacherous threatened

The passoc values from Tables 8 and 9 were then treated exactly analogously to the relative frequencies used in the preceding sections, i.e. the formulas in (3), (4), and (5) were applied. Results. The results of the procedure described above are shown in Table 10. Again, for each of the three versions of the overlap model, the overlapping adjectival collocates are listed in decreasing order of combined probability. Table 10. Association-based overlap between [men are ADJ] and [wolves are ADJ] Symmetrical Rank 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22.

Adjective dangerous intelligent good at smart scary strong numerous afraid sensitive noble useless bad aggressive restless tall stupid free powerful brave wise satisfied active

Source-dominant

Target-dominant

pcombined Rank Adjective pcombined Rank Adjective 4.01E–04 1. dangerous 2.45E–03 1. intelligent 3.01E–04 2. good at 2.21E–03 2. dangerous 1.58E–04 3. intelligent 1.47E–03 3. scary 9.73E–05 4. smart 8.60E–04 4. smart 8.85E–05 5. scary 7.23E–04 5. good at 5.28E–05 6. strong 5.53E–04 6. afraid 4.26E–05 7. numerous 5.06E–04 7. strong 3.14E–05 8. useless 4.13E–04 8. numerous 2.34E–05 9. sensitive 3.65E–04 9. noble 2.12E–05 10. tall 3.03E–04 10. bad 2.11E–05 11. free 2.88E–04 11. sensitive 1.68E–05 12. noble 2.77E–04 12. aggressive 1.66E–05 13. stupid 2.50E–04 13. useless 1.41E–05 14. restless 2.49E–04 14. restless 1.13E–05 15. afraid 2.44E–04 15. powerful 1.06E–05 16. aggressive 2.31E–04 16. stupid 1.04E–05 17. bad 2.21E–04 17. brave 8.33E-06 18. brave 1.58E–04 18. tall 7.59E-06 19. wise 1.33E–04 19. common 6.09E-06 20. powerful 1.19E–04 20. free 4.94E-06 21. satisfied 1.13E–04 21. active 4.76E-06 22. active 9.00E–05 22. wise

pcombined 3.55E–03 3.27E–03 1.15E–03 1.12E–03 8.98E–04 7.19E–04 6.93E–04 5.50E–04 3.54E–04 3.12E–04 3.10E–04 2.94E–04 2.34E–04 2.12E–04 2.02E–04 1.38E–04 1.32E–04 1.26E–04 1.16E–04 1.16E–04 1.15E–04 1.13E–04


Table 10. (cond.) Symmetrical

Source-dominant

Target-dominant

Rank Adjective

pcombined

Rank Adjective

pcombined

Rank Adjective

pcombined

23. 24. 25. 26. 27. 28. 29.

3.28E-06 2.48E-06 2.16E-06 2.08E-06 1.50E–06 7.70E-07 6.00E-07

23. 24. 25. 26. 27. 28. 29.

8.77E–05 6.24E–05 6.12E–05 5.12E–05 4.08E–05 2.91E–05 2.45E–05

23. 24. 25. 26. 27. 28. 29.

9.69E-05 7.35E–05 5.09E-05 4.45E–05 3.00E–05 2.32E–05 1.90E–05

common dead quiet similar close nice present

dead quiet close common similar nice present

satisfied similar quiet dead close nice present

Discussion. As with the frequency-based method, the results are encouraging for all three versions, which rank some of the most relevant adjectives among the top ten overlapping collocates, e.g. intelligent, dangerous, scary, strong, and, in the case of the target-domain dominant version, bad. The latter version again seems best, not only because it includes bad, which has disappeared from the top ten in the other two versions, but also because it ranks the highly relevant scary highest. Note also, that some positive ‘lone wolf ’ attributes are again mixed in with the negative ones (noble, sensitive, restless, free). Crucially, the problematic adjective good has disappeared completely from the list.

3.2 Testing the model Aims. In this section, I will once again apply the model developed above to other source domain items in order to test its generalizability. Data and Method. The same data as in Section 1 were used. The method used was the same as that just demonstrated for wolf. Table 11. Strongest adjectival collocates in the pattern [gorillas are ADJ] Rank Adjective 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.

Rank Adjective

passoc

endangered 1.36E-01

passoc

11.

strong

2.04E–02 21.

threatened shy gentle large peaceful intelligent vegetarian sensitive planteating

12. 13. 14. 15. 16. 17. 18. 19. 20.

big protected dangerous sociable vegetarians monogamous polygynous habituated folivorous

1.77E-02 1.60E-02 1.55E-02 1.43E-02 1.43E-02 1.43E-02 1.43E-02 1.43E-02 1.43E-02

4.58E-02 3.75E-02 3.48E-02 2.98E-02 2.82E-02 2.74E–02 2.42E-02 2.29E-02 2.23E-02

Rank Adjective

22. 23. 24. 25. 26. 27. 28.

familyoriented smart cute ferocious stupid susceptible fierce aggressive

passoc 1.43E-02 1.25E-02 1.17E-02 1.17E-02 1.00E-02 9.70E–03 7.60E–03 7.40E–03


Table 12. Association-based overlap between [men are ADJ] and [gorillas are ADJ] Symmetrical Rank 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20.

Adjective dangerous intelligent sensitive strong reluctant smart susceptible stupid scary insensitive capable aggressive protective close intellectual quiet powerful active nice present

Source-dominant pcombined 2.33E–04 1.97E–04 1.31E–04 1.18E–04 1.03E–04 9.50E–05 9.31E–05 5.90E–05 3.48E–05 2.92E–05 2.60E–05 2.37E–05 1.33E–05 1.23E–05 8.70E–06 8.28E-06 7.48E-06 7.31E-06 1.32E-06 1.10E–06

Rank 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20.

Adjective dangerous reluctant intelligent susceptible sensitive smart strong stupid scary insensitive capable aggressive protective close intellectual quiet powerful active nice present

Target-dominant pcombined 1.87E–03 1.82E–03 1.19E–03 9.45E–04 8.63E–04 8.50E–04 8.28E–04 5.90E–04 4.53E–04 3.97E–04 3.61E–04 2.75E–04 2.19E–04 1.75E–04 1.59E–04 1.22E–04 1.13E–04 1.11E–04 3.81E–05 3.32E–05

Rank 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20.

Adjective intelligent dangerous sensitive strong smart susceptible stupid reluctant scary aggressive insensitive capable close protective quiet powerful active intellectual nice present

pcombined 2.32E–03 1.90E–03 1.73E–03 1.55E–03 1.09E–03 9.50E–04 7.68E–04 5.74E–04 4.53E–04 4.19E–04 3.97E–04 3.68E–04 2.45E–04 2.22E–04 1.95E–04 1.81E–04 1.77E–04 1.62E–04 3.98E–05 3.48E–05

Results. The strongest adjectival collocates for the pattern [gorillas are ADJ] are shown in Table 11. Unlike in the case of the frequency-based analysis, some relevant adjectives for the metaphorical interpretation of gorilla now occur among the top twenty adjectival collocates, but not in very high positions. Table 12 shows the overlapping collocates of [men are ADJ] and [gorillas are ADJ]. As with the frequency-based method, the results are promising if compared to the individual collocates for gorilla. Many relevant adjectives appear among the top ten: dangerous, strong, stupid, scary, insensitive (in the symmetrical and sourcedominant versions) and aggressive (in the target-dominant version). However, again the results are less encouraging than for wolves because a number of incorrect adjectives occupy high positions, most importantly, intelligent, smart, and sensitive. It seems that the association-based method was not able to remove these as it removed good for wolves. It is difficult to distinguish between the different versions. If anything, we can say that the target-domain dominant version is slightly worse since it contains the incorrect intelligent in first position. Next, the most strongly associated adjectives in the pattern [pigs are ADJ] are shown in Table 13. As before, these individual collocates would not lead to the right metaphorical interpretation of pig. Table 14 shows the overlapping collocates for [men are ADJ] and [pigs are ADJ]The results are similarly mixed as they were in the


Table 13. Strongest adjectival collocates in the pattern [pigs are ADJ] Rank Adjective

passoc

Rank Adjective

passoc

Rank Adjective

passoc

1. 2. 3. 4. 5. 6. 7. 8.

6.58E-02 4.64E–02 4.10E-02 2.97E-02 2.91E-02 2.40E-02 2.20E-02 2.13E-02

9. 10. 11. 12. 13. 14. 15. 16.

2.12E-02 1.70E-02 1.70E-02 1.70E-02 1.51E-02 1.43E-02 1.39E-02 1.32E-02

17. 18. 19. 20. 21. 22. 23.

1.32E-02 1.32E-02 1.25E-02 1.25E-02 1.16E-02 9.70E–03 8.20E–03

susceptible intelligent clean fat pink dirty cute healthy

smart spoilt important same inquisitive aggressive equal hardy

prolific clever plump greedy filthy ugly able

case of the frequency-based analysis. The rankings are in some ways an improvement over the individual collocates of pigs are ADJ (for example, the counterintuitive clean is removed from the list), in other ways they are worse (the highly relevant fat and dirty occur in lower positions). Overall, the results are slightly better than those for the frequency-based model: the relevant adjectives fat and dirty are ranked higher by the association-based model in all three versions. However, in both cases, there are a number of irrelevant or counterintuitive adjectives among the top-ranked collocates. As discussed in Section 1, the metaphorical interpretation of pig is less homogeneous than that of the other words treated here; this may be a reason for the somewhat unconvincing results of both the frequency-based and the association-based model. Discussion. Overall, we can say that the association-based model offers a slight improvement over the frequency-based model: in all cases, some irrelevant or counterintuitive adjectives were removed or demoted by the model. However, a number of irrelevant or counterintuitive collocates remain. There are three possible reasons for this. First, the association-based model may not be sufficiently sophisticated. Future research could test different statistical methods to determine whether some are better than others. However, I would venture the guess that this is not the main problem. Second, the association strengths may have been calculated on an insufficiently comparable basis, i.e. the base frequencies derived from the BNC may not reflect the base frequencies on the Internet closely enough. The obvious solution would be to use a different corpus (which is currently impossible, since no suitable corpora are available) or to use Internet-derived frequencies as a baseline (which is also impossible, since there is no straightforward way of identifying, retrieving and counting all adjectives from the Internet). However, advances in natural language processing and the availability of larger annotated corpora may help to solve these problems within the next few years. Third, the problem of unrepresentative text types, already mentioned in connection with the frequency-based model exists for the associationbased model just as it does for the frequency-based model. However, the advances just mentioned will also help to solve this problem at some point in the near future.


Table 14. Association-based overlap between [men are ADJ] and [pigs are ADJ] Symmetrical Rank 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36.

Adjective susceptible intelligent smart happy same healthy equal fat dirty able lazy aggressive clever dangerous aware messy responsive lean sensitive noble close capable lost active free present cold funny common heavy bad difficult nice available similar young

Source-dominant pcombined 6.32E–04 3.34E–04 1.61E–04 1.34E–04 1.16E–04 1.15E–04 1.06E–04 8.02E–05 6.72E–05 6.40E–05 4.74E–05 4.58E–05 3.83E–05 3.75E–05 3.69E-05 3.11E–05 2.53E–05 2.33E–05 1.77E–05 1.62E–05 1.53E–05 1.35E–05 1.04E–05 9.01E-06 8.80E–06 6.20E–06 5.94E-06 5.88E-06 4.40E–06 2.34E-06 2.32E-06 1.80E–06 1.65E-06 1.10E–06 9.60E-07 4.80E-07

Rank 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36.

Adjective susceptible happy intelligent smart equal aware same healthy dangerous able lazy fat dirty messy aggressive responsive clever lean sensitive free capable noble close lost cold active funny bad present common difficult heavy nice similar available young

Target-dominant pcombined 2.46E–03 1.64E–03 1.55E–03 1.11E–03 8.96E–04 8.95E–04 8.87E–04 7.88E–04 7.50E–04 7.06E–04 6.64E–04 4.65E–04 4.34E–04 3.98E–04 3.83E–04 3.41E–04 3.33E–04 3.20E–04 3.17E–04 2.65E–04 2.60E–04 2.41E–04 1.95E–04 1.74E–04 1.40E–04 1.24E–04 1.11E–04 8.20E–05 7.87E–05 5.93E–05 5.69E-05 5.52E–05 4.26E–05 2.77E–05 2.35E–05 1.96E–05

Rank 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21. 22. 23. 24. 25. 26. 27. 28. 29. 30. 31. 32. 33. 34. 35. 36.

Adjective susceptible intelligent smart healthy fat same dirty equal happy aggressive able clever lazy messy responsive lean dangerous close noble aware sensitive active present lost capable common funny cold free heavy nice available bad difficult similar young

pcombined 6.45E–03 3.94E–03 1.85E–03 1.57E–03 1.54E–03 1.40E–03 1.27E–03 1.21E–03 9.48E–04 8.09E–04 7.24E–04 7.11E–04 4.92E–04 4.36E–04 3.73E–04 3.52E–04 3.06E–04 3.05E–04 2.70E–04 2.50E–04 2.34E–04 2.19E–04 1.96E–04 1.94E–04 1.91E–04 1.56E–04 1.28E–04 1.03E–04 9.84E–05 6.49E-05 4.97E–05 4.92E–05 4.31E–05 4.24E–05 3.39E-05 1.70E–05

4. Evaluation of the model against native-speaker interpretations Aims. In introducing the overlap model, I was deliberately vague as to the nature of the language processor. The collocational overlap model of metaphor interpretation would, in my view, be useful even if it were merely interpreted as a suggestion for the unsupervised deduction of metaphorical interpretations by a machine-


learning system, with no claims of psychological reality for human speakers. However, the model would, of course, be more interesting from a theoretical perspective if it could be shown that it is psychologically real, i.e. that collocational overlap in fact does contribute to the interpretation of metaphorical expressions by human speakers. Even if we were to limit ourselves to an applied machine-learning interpretation of the overlap model, we would still have to ensure that it actually identifies the same attributes as would human speakers. Therefore, I decided to perform a simple elicitation study confronting native speakers with the three metaphorical statements discussed above in order to evaluate the performance of the overlap model in predicting native-speaker interpretations of conventionalized metaphors. I also included one additional statement exemplifying a non-conventionalized metaphor in order to test whether the model could predict the interpretation of novel metaphorical statements. The non-conventionalized metaphorical statement was That man is a dolphin. This statement was constructed to be maximally parallel to the other three in that it also involves a target concept from the domain of animals. Dolphin was chosen because (i) there is no conventionalized interpretation attached to it, (ii) dolphins are popular animals and hence likely to be familiar enough to native speakers of English to have stereotyical attributes associated with them, and (iii), dolphins are talked about frequently enough to allow a corpus analysis analogous to those performed above. Data and methods. There seem to be at least two possible ways of testing the results of the corpus analysis against native-speaker interpretations. The first, and perhaps best way would be to confront speakers with statements such as Richard is a wolf (conventionalized) or Sam is a dolphin (non-conventionalized) and ask them to list properties that they would associate with Richard and Sam. These lists could then be concatenated and compared to the frequency lists of adjectives associated with man and wolf/dolphin, as well as the lists yielded by the various versions of the collocational overlap model. However, especially with novel metaphors, this would require a large number of native-speaker responses to ensure the reliability of the patterns that would – hopefully – emerge. The second possibility would be to confront speakers with the lists resulting from the corpus analysis and ask them to choose the list(s) that fit best their interpretation of these sentences. Here, the number of choices could be limited and thus focused comparisons between different lists could be elicited even with a comparatively small number of native speakers. Since I only had 12 native speakers of English available, I settled on the second possibility. Due to the limited number of native speakers available, it was clear that I could test two conditions at most before the subjects would no longer approach the data on the basis of their spontaneous native-speaker interpretations. I therefore decided on the following two comparisons:


Condition 1: SD-node frequency list (man) vs. TD-node frequency list (wolf, gorilla, pig, dolphin) vs. Frequency-based symmetrical overlap. Condition 2: Frequency-based symmetrical overlap vs. Association-based symmetrical overlap. The first condition was intended to test whether the overlap model would outperform the simple source-domain frequency model. The second condition was meant to test whether the association-based overlap model would outperform the frequency-based one. These questions seem more fundamental than the issue of distinguishing between the source-dominant and the target-dominant versions. Only the top five adjectives of the respective lists were used in the experiment. This limitation is rather strict in view of the discussion of the corpus results above, which shows that important adjectives as well as important differences between the lists often occur a little further down the list. However, the limitation was necessary to ensure that the task would not become too complex for the subjects. For the three sentences discussed in the preceding sections, the lists for the two conditions can be read off the tables presented in the preceding sections and will not be repeated here. For the sentence That man is a dolphin, the lists are shown in Table 15. The subjects were 12 exchange students from English-speaking countries at the University of Bremen (six from the United States, five from Britain, and one from Ireland). They were tested individually or in small groups. Subjects sat at desks facing a large computer screen. Before the experiment began, they were given the following instruction: You will see a series of statements of the form “That man is a giant.” You will then be presented with a number of lists naming properties. In each case, choose the list that contains the properties most relevant to your interpretation of the sentence. If more than one list contains the same relevant property, choose the list on which that property occurs closer to the top of the list. The lists for each condition were then shown side by side in random order on the computer screen using Table 15. Adjective lists for That man is a dolphin Simple frequenciesa Rank 1. 2. 3. 4. 5. a

Adjective intelligent common social able large

Frequency overlapb p 8.53e-02 8.19e-02 3.09e-02 2.75e-02 2.41e-02

Rank 1. 2. 3. 4. 5.

Adjective intelligent happy able free common

Association overlapc p 5.94E–04 5.71E–04 5.71E–04 2.97E–04 2.85E–04

Rank 1. 2. 3. 4. 5.

Most frequent adjectives in the frame [dolphins are ADJ] Frequency-based overlap between [men are ADJ] vs. [dolphins are ADJ] c Association-based overlap between [men are ADJ] vs. [dolphins are ADJ] b

Adjective intelligent happy smart able free

p 1.07E–03 2.76E–04 2.67E–04 9.83E–05 6.56E–05


a white font against a black background. Each list was enclosed by a box and had an identification number appearing directly below the list. At the top of each screen, the relevant sentence was displayed. Subjects then recorded their responses on a paper questionnaire by checking boxes corresponding to the identification number of their chosen list. Results. The results for the first condition are shown in Table 16. In the case of wolf, there is a clear preference for the frequency-based overlap model compared to the two simple frequency lists. This difference is very significant (χ2=11.64 (df=2), p ), or fragments of verbal morphology become reanalyzed as question particles (e.g. Québecois ti). In this sense, question particles tend to usurp the interrogative strategy that before was basically intonational. The same holds for classical tag questions, which are usually marked for an additional intonational pattern not on the declarative

256 Wolfgang Schulze

segment, but on the tag. It may be safely hypothesized that interrogative morphology within verbs once had an analogous background such as North Slope Iñupiaq imiq-pa-t (drink-q-3pl:s) ‘Are they drinking?’ (MacLean 1986: 63)). Interrogative morphemes may share with the corresponding particles the function of focusing the preceding lexical segment., Compare the following examples from Hunzib (East Caucasian, van den Berg 1995: 115; Q = question morpheme, $ = second part of discontinuous lexeme):

(5) a. cax-čó-y (Hunzib) write-pres:i/ii-q ‘Does (s)he writeIs (s)he writing?’ [I do not know.] b. ča-y-x-čo write-q-$-pres:i/ii ‘Does (s)he writeIs (s)he writing? [‘Is it really so that (s)he writesis writing? (S)he shouldn’t be!] c. y-á-y-č’el-er? ii-freeze-q-$-past ‘[Did you say] did you freeze/were you freezing?’ [I did not get you.]

Tagging techniques are widespread among the languages of the world. They may include words for or <no> or other, verificational particles, clefts or partial clefts (i.e. full verbal or partial (identificational) or generic echoes) or an alternative phrasing (or-anacolutha). If the tag is lexically transparent (that is not fully grammaticalized), it is the tag that represents the interrogative construction, not the proposition itself, which remains affirmative. Identificational tags often have an additional verificational value (e.g. French est-ce que). Clearly, clefted tags come close to clause-external focusing strategies: They focus either on the properties of the verbal relation ( so-called propositional focus) or on individual constituents. In the latter case, polar questions differ from constituent questions only in that they identify the potential conceptual domain (often stressed by intonational patterns), whereas wh-terms render it indefinite. In fact, polar questions do not constitute a class of question constructions different from the set of constituent questions, but merely a semantically more transparent way of construing questions. Note that no language seems to have a strategy to question a verbal relation itself (in terms of constituent questions). In other words, there are no pro-verbs that would ask for nothing but the verb. Even so-called rogoverbs (Hagège 2003) are usually marked for a complex conceptual structure that entails a wh-segment and a more or less overt verbal dummy (Wh-Verb, Wh:Verb, Wh:Ø). This observation correlates with the hypothesis that any world stimulus has to be processed by somehow identifying its relational segment, whereas referential constituents can be bleached to uncertainty.

Towards a cognitive typology of questions 257

A noteworthy way of construing polar questions is given by the so-called syntactic inversion (or changes in word order). This technique is typical of a number of Germanic languages, but is also present in some Romance and Slavic languages. In addition, Haspelmath et al. (2005: 470) mention two Austronesian and one Arawakan language which also show word order variation (König and Siemund (2006) add Finnish and Malay). Most importantly, all languages at issue mark the interrogative construction by verb-initial word order. Here, I cannot go into the details of this type of syntactic marking. Still, it should be noted that there is sufficient evidence to claim that, at least in Germanic, the word order of polar questions does not seem to be a secondary change, but rather the reflex of the original word order marked for an interrogative intonational pattern. The hypothesis put forward here suggests that Germanic was once marked for a V-initial typology that later became reanalyzed as a question construction after the prefield had been opened to the focus (topic) function. As has been pointed out, constituent questions differ from polar questions with respect to the degree of lexical and conceptual specification of a constituent. Usually, constituents questioned in an interrogative construction are represented by some kind of interrogative pronoun (note that the term ‘pronoun’ also often applies to those pro-forms which in fact add a relational component, such as how, etc.). The fact that wh-constituents focus on a specific knowledge state of the speaker (better: on knowledge deficiency) means that such constituents often conform to the focal patterns of a given language. In addition, there seems to be an intimate relation between the verbal (relational) domain and a wh-term: Quite often, a wh-term precedes its verb. In sum, we can posit the following correlation:

(6) Assertive FOC V X X > X FOC V X > X X FOC V > X X V FOC >

Interrogative (constituent questions) wh V X X X wh V X X X wh V X X V wh (?)

Postponed wh-terms are rare and often coupled with a specific pragmatic notion, as seen in German and Arbore (Western branch of Omo-Tana, Makro-Somali, Hayward 1984: 116, transcription simplified):

(7) Wer komm-t? > who:nom come:pres-3sg:pres Es komm-t wér? foc come:pres-3sg:pres who:nom ‘Who is coming?’


(8) móha yéečče máh person come:3sg:past who ‘Who came?’

Obviously, such strategies involve the splitting of the focal domain with the help of a dummy referent (German es . . . wer? Arbore móha . . . máh?) in order to open a slot for an afterthought.

3.3 Uncertainty Constituent questions are marked for a presupposition that is related to the givenness of a questioned referent or reference-relation cluster. By this I mean that a question like Who will come tomorrow? entails the assumption that there is ‘someone’ who ‘will come.’ This presupposition conditions that the wh-domain can be subcategorized to the extent that a language system allows, for instance English who/what or Tabasaran (East Caucasian) fuž (who:sg), fužur (who:pl), fi (what:sg), fiyar (what:pl). The same holds for the relational marking of wh-terms, e.g. case, or for class/gender subcategorization. An example from Hunzib (East Caucasian), a noun-classifying language, is:

‘Who (a man) came?’ (Class I) (9) suk’u ãq’e-r who I:come-past suk’u y-ãq’e-r ‘Who (a woman) came’? (Class II) suk’u n-ãq’e-r ‘Who (a child) came?’ (Class V)

The fact that wh-terms can be subcategorized suggests that they have bleached referential properties. In fact, the well-known debate concerning the relationship between wh-terms and indefinite pronouns illustrates that referential ‘indefiniteness’ is a crucial segment with respect to the conceptual layer of wh-terms. For instance, in Chiricahua, the wh-term <what> is derived from iyáa ‘something’ (> iyáa-hee). In Coptic (Afroasiatic), wh-terms are also used to indicate referential vagueness (nim ‘someone, the so-and-so; who’; aš ‘what(soever), thing; what’ (Till 1966). Although we do not yet have at hand a full coverage of grammaticalization paths that would illustrate the secondary nature of wh-terms, it can said with some certainty that a likely source for wh-terms may have been words for , , , etc.; see Udihe ni ‘who’ ~ ninta ‘man’, Hunzib suk’u ‘who’ = ‘man’ (with different inflectional patterns), šiyo ‘what’ ~ šiyon ‘some, a little’, Arbore móh ‘man’ vs. mah ‘who,’ Idg. *kwe-/*kwo- etc.: *-s , *-d vs. *kwe- etc. ‘some,’ ‘any’ (Latin -que), Korean nugu ‘who’ ~ ‘anyone’, Kilmeri bo ‘what’ = ‘word,’ Udi šu ‘who’ ~ išu ‘man,’ German wer ‘who’ ~ ‘somebody’ etc. Even though indefinite pronouns may again be derived from interrogatives


(compare Russian kto ‘who’ > kto-nibud’ ‘anybody,’ Udi šu ‘who’ > šuk’al ‘anybody’ etc.), it seems reasonable to assume that we have to start from prototypical ‘dummy’ nouns that occur as wh-terms under intonational patterns, but as indefinite pronouns with assertions. (10) summarizes the possible grammaticalization paths: (10)

DUMMY REFERENCE [−Q]

[+Q]

INDEF I

Wh

Construction O + DUMMY

Lexical DUMMY

INDEF II

In sum, it emerges that constituent questions do not represent a constructional pattern distinct from that of polar questions. Both share the basic property of intonation and only differ in the degree to which referentiality is concretized. In other words, a Cognitive Typology has to start from this aspect, which is obviously the most general pattern.

4. Interrogation and memory mismatch In Section 3, I tried to show that interrogative marking is characterized by a relatively high degree of formal variation. Most of the relevant variants result from the habitualization of interrogative styles that are based on focusing strategies. This holds both for polar and constituent questions. In addition, it should be noted that — as has been observed above — questions do not reflect an immediate conceptual reaction to a given world stimulus, but result from the correlation of an external or internal world stimulus with a corresponding memory segment. Hence, the type of cognitive attention towards the experience of a world stimulus as embodied in the attention flow differs from assertive reactional types in that the stimulus does not find adequate or sufficient correspondences in memory. This experienced deficiency can again result in a number of different linguistic strategies, both within a language and cross-linguistically.


Hence, a Cognitive Typology-based approach to questions has to start from the underlying universal dihairetic patterns of experiencing a world stimulus. Nevertheless, it also has to consider typological variation, which stems (among others) from types of specification, metaphorization, and categorization that manipulate these patterns. We can start from the hypothesis that intonation represents the basic pattern of indicating a memory mismatch in terms of a linguistic utterance and ask why intonation seems to be the most natural way to react upon such a memory mismatch. Particles and morphological means to mark questions echo this basic strategy, sometimes modifying the semantics of pitch variation due to the semantics underlying these means. Or, to put it differently, particles and morphological and syntactic means of question marking are more language-related strategies to articulate a (more or less strong) µ/α-mismatch. The most pronounced way to encode this mismatch is to link a positive assertion with its negative mode, as in the Mandarin Chinese example quoted by König and Siemund (2006): (11) tā zài jiā bu zài jiā? 3sg at home neg at home ‘Is s/he at home?’ But note that again, this constructional type (µ/α-match () µ/α-mismatch) is accompanied by pitch variation. Other question particles have a different origin; still it can safely be assumed that they, too, once somehow encoded the conceptual representation of a µ/α-mismatch experience. In terms of Radical Experientialism, intonation is derived from pre-linguistic schemata that relate variation in pitch to degrees of certainty. The underlying motoric schema is also present in gesture; see the raising of the arms or of the shoulders to indicate uncertainty or even interrogativity. The same holds for the raising of the eyebrows or of the head. Obviously, we have to deal with a motor schema that is strongly coupled with notions of and (in an embodied sense). The indication of ‘openness’ (in whatever way) seems to be an immediate reactional type with respect to memory mismatch. Here, I will disregard the further elaboration of the question as to why ‘openness’ (resulting from bodily ‘upness’) qualifies as a reactional pattern for a given memory mismatch. Turning to linguistics, it becomes clear that an iconic representation of ‘upness’ is pitch variation (> up). Hence, pitch raising corresponds to the above-mentioned non-linguistic reactional types. From this, we can assume that prototypical questions are marked for pitch. If we include the reactional type of the hearer, we can hypothesize that a question marked for pitch raising serves as a stimulus for the hearer to close the prosodic pattern: high > not-terminated > open > stimulus(!!close). The relevant prosodic pattern can be described as in Figure 1. Here, the term AIF (Attention Information Flow, see Schulze 2004) refers to the ensemble of linearization patterns that are re-


High

?

AIF

Low Stimulus

!

AIF Reaction

Figure 1.

lated to both the attention of cognition towards the incoming world stimulus and the informational attitude towards a hearer. Accordingly, a speaker may additionally pinpoint a domain with the experienced world stimulus (ws μ>α) that becomes specifically marked for uncertainty during the dihairesis process. As a result, an additional constituent focus can occur. Prototypically, the intonational focus on a constituent in a question emphasizes the assumed givenness of the referent represented by the constituent (e.g. Did MARY give him the book?). Focus-like properties can also be described for polar questions marked by specific question markers. For instance, Russian li not only marks interrogativity itself, but also focuses on the constituent that precedes this marker. The same holds e.g. for Turkish mı can term such elements ‘floating question markers.’ Non-floating (in situ) markers usually do not focus on a constituent but on the whole construction used to encode an uncertain ws μ>α, often in terms of a tag question. Most importantly, a clause-initial, often verificational tag (French estce que) does not have an interrogative intonational pattern itself when it is strongly grammaticalized. The same holds for a number of clause-final tags reduced to question markers, especially if they stem from -anacolutha. As has been said above, constituency is strongly related to the type of linearization applied to the processing of a world stimulus. The more explicit and detailed linearization is, the more a world stimulus is processed in terms of individual features. However, this type of elaborated dihairesis makes the mirroring process more complex: The cognition now activates a more detailed memory segment, leading to the possibility that some parts of it match a world stimulus better than others. The higher resolution of a ws (see Schulze in press b) thus means that a part of the dihairetic output has to be represented by vague conceptualization. As I argued in Section 3, the lexical encoding of vagueness is directly related to dummy concepts, which may again show up as interrogative pronouns. As a result, constituent questions emerge. Quite often, they are marked for additional strategies that stem from the general patterns described above, that is, from intonation and focus. Naturally, it would be of great explanatory power if we had a cross-linguistic study that would list the preferred target of wh-strategies in actual texts from a cross-linguistic per-


Table 1. Core actants Modal Locative Cause Attributive Temporal Telic

(wer/was) (wie) (wo-) (warum) (welch-) (wann) (wozu)

113 92 39 38 22 14 5

34.98% 28.48% 12.07% 11.76% 6.81% 4.33% 1.54%

spective. Here, I cannot fill this gap. Still, a cursory glance at a sample text from German (Karl May: Kinder des Herzogs, pp.6–109; 20,500 tokens) which contains extensive dialogue sequences illustrates that core actants and modal variations within the verbal relation are the most frequent sequences marked for wh-constituents; see Table 1. From this, we may assume that memory mismatch in terms of uncertainty is most often related to the insufficient identification of coreferents within the dihairesis process.

5. Conclusions In my chapter, I have argued that the world of interrogative structures is directly related to cognitive procedures for processing a world stimulus. Accordingly, the meaning of interrogative constructions reflects a state of cognition that is marked for a memory mismatch. Most importantly, the major formal types of interrogative constructions can be derived from schematic processes that are active during the dihairesis of a world stimulus. Questions of the kind discussed above differ from speech act verb constructions that involve some kind of interrogative mode. For instance, utterances like I am asking you whether you will come or Tell me whom you have seen, etc. are communicative in nature because they include a speech act verb (ask, tell) and the encoding of the addressee. Simple questions, however, often lack this communicative component, unless a second person is overtly marked, as in Are you going to the market? Still, even in these instances, we may question the status of you (the addressee) as a partner in communication. There is sufficient evidence that you represents a rather vague conceptualization reflecting a role swap: ego conceptualizes tu (the conceptual basis for you) as some kind of alter-ego (see Schulze 1998, 2001). Role swapping allows ego to map properties of its own cognition onto another individual (mixed with hypotheses about the mental state of the addressee). I assume that questions are part of these role swapping strategies. Accordingly, questions are not primarily structured in terms of information-seeking devices, as it would be formulated according to an optimistic communication theory. They do not invite a hearer to answer the question, but represent reactional patterns that


reflect the given state of cognition. In this sense, interrogatives are nothing but a variant of negation (which explains why negative concepts and their lexical representations often occur in the context of interrogatives). Unlike negation, however, questions do not cancel a μ/α-match but keep this match vague. Schematically, the whole process of interrogation can be described as follows: An external or internal world stimulus tentatively mirrored in memory results in a contortion that overstretches the μ/α-relation or (in part) blocks the matching process. This process leads to a state of cognition that is marked for an ‘open’ μ/α-hypothesis (?ws α). The openness of this state evokes a secondary hypothesis of ego, namely that an alterego would interpret the articulation of ?wsα in terms of an imitation process. alterego would then process the world stimulus (or its construction) in an analogous way and articulate the degree of μ/α-match according to non- ego’s general cognitive setting. It should be noted that the projection of ego’s cognitive state onto an alter-ego is not an original property of question strategies. Rather, we have to deal with the assimilation of secondary cognitive procedures stemming from the internalization of communicative routines in a social group. Naturally, this secondary self-justifying strategy of a cognitive system can also have its linguistic realization. This assumption goes together with the hypothesis that communication itself is a social construct of human beings. Accordingly, human beings have since long habitualized their linguistic practice as communication, a process that has had its impact (by recursion) on individual linguistic knowledge itself. However, the corresponding techniques do not represent the original constructional devices. These are direct or indirect representations of a memory mismatch, regardless of whether there is a potential hearer present in the speech act situation or not.

References Austin, John L. 1962. How to Do Things with Words. Oxford: Clarendon Press. Bussmann, Hadumod. 1996. Routledge Dictionary of Language and Linguistics (Translated and edited by Gregory Trauth and Kerstin Kazzazi). London: Routledge. Diessel, Holger 2003. The relationship between demonstratives and interrogatives. Studies in Language 27(3): 635–55. Dik, Simon C. 1997. The Theory of Functional Grammar, Part 2: Complex and Derived Constructions (Edited by Kees Hengeveld). Berlin: Mouton de Gruyter. Grice, Paul. 1975. Logic and conversation. Speech Acts [Syntax and Semantics 3], Peter Cole and Jerry L. Morgan (eds) 41–58. New York: Academic Press. Enghahl, Elisabet. 1986. Constituent Questions. Dordrecht: Kluwer. Evans, Vyvyan (nd). Lexical concepts, cognitive models and meaning-construction. http://www. sussex.ac.uk/Users/vyv/LexicalConceptsCognitiveModels&Meaning-Construction.pdf


Fodor, Jerry. 1974. Special sciences – The disunity of science as a working hypothesis. Synthesis 28: 77–115. Ginzburg, Jonathan. 2002. Questions, Queries, and Facts: A Semantics and Pragmatics for Interrogatives. Cambridge: Cambridge University Press. Graesser, Arthur C. 1994. Question asking and answering. Handbook of Psycholinguistics, Morton Anne Gernsbacher (ed.), 517–38. San Diego: Academic Press. Hagège, Claude. 2003. A paradox in linguistic theory: Rogoverbs, or WHATed we to interrogative verbs? Paper presented at the biennial meeting of the Association of Linguistic Typology Cagliari 2003. Haspelmath, Martin, Dryer, Matthew S., Gil, David and Comrie, Bernard (eds). 2005. The World Atlas of Language Structures. Oxford: Oxford University Press. Hayward, Dick. 1984. The Arbore Language: A First Investigation. Hamburg: Buske. Heine, Bernd and Kuteva, Tania. 2002. World Lexicon of Grammaticalization. Cambridge: Cambridge University Press. Hintikka, Jaakko. 1978. Answers to questions. In Questions, Henry Hiż (ed.), 279–300. Dordrecht: Reidel. Hiż, Henry (ed.). 1978. Questions. Dordrecht: Reidel. Hudson, Richard A. 1975. The meaning of questions. Language 51:1–31. Keenan, Edward O. 1976. On the universality of conversational implicature. Language and Society 5: 67–80. König, Ekkehard and Siemund, Peter. 2007. Speech act distinctions in grammar. In Language Typology and Syntactic Description, Vol. 1, Timothy Shopen (ed.). Cambridge: Cambridge University Press. Lakoff, George 1987. Women, Fire, and Dangerous Things. Chicago: University of Chicago Press. MacLean, Edna Ahgeak. 1986. North Slope Iñupiaq Grammar: First Year (3rd ed. revised). Fairbanks Alaska Native Language Center. Meyer, Michael (ed.). 1988. Questions and Questioning. Berlin: Walter de Gruyter. Schulze, Wolfgang 1998. Person, Klasse, Kongruenz. Vol. 1 (in two parts). Munich: Lincom. Schulze, Wolfgang 2001. Selbstlernen und Selbstreflexion: Grundlagen einer Emergenz-Theorie der sprachlichen Interaktion. Munich Working Papers in Cognitive Typology 1. Schulze, Wolfgang 2004. Pragmasyntax: Towards a cognitive typology of the attention information flow in Udi narratives. In Linguagem, Cultura e Cognição: Estudos de Linguística Cognitiva, Augusto Soares da Silva, Amadeu Torres and Miguel Gonçalves (eds), 545–74. Coimbra: Almedina. Schulze, Wolfgang. In press a. Sprache als kommunizierte Wahrnehmung. Paris (FS contribution). Schulze, Wolfgang. In press b. A new model of metaphorization: Case semantics in East Caucasian. In Metonymy and Metaphor in Grammar, Antonio Barcelona, Klaus-Uwe Panther, and Linda Thornburg (eds). Amsterdam: John Benjamins. Searle, John R. 1969. Speech Act: An Essay in the Philosophy of Language. Cambridge: Cambridge University Press. Siemund, Peter. 2001. Interrogative constructions. In Language Typology and Language Universals, Vol. 2, Martin Haspelmath, Ekkehard König, Wulf Oesterreicher and Wolfgang Raible (eds), 1010–1028. Berlin: de Gruyter. Till, Walter C. 1966. Koptische Grammatik (Saidischer Dialekt). Leipzig: Verlag Enzyklopädie. Van den Berg, Helma. 1995. A Grammar of Hunzib. Munich: Lincom.

chapter 13

Brutal Brits and persuasive Americans Variety-specific meaning construction in the into-causative Stefanie Wulff, Anatol Stefanowitsch, and Stefan Th. Gries University of Bremen/University of Bremen/ University of California, Santa Barbara

“England and America are two nations divided by a common language.” (George Bernard Shaw) Adopting a construction-based view of language (Goldberg 1995), we demonstrate that it is possible to uncover differences between British and American English at the lexicosyntactic level, showing that the collexemes, i.e. the words significantly associated with a construction, are variety-dependent. To this end, we compare more than 5,000 verb pair types as they occur in the two varieties in the so-called into-causative construction (as in He tricked me into employing him) and submit them to the scrutiny of a statistical test called distinctive collexeme analysis, which identifies those verbs that distinguish best between the two varieties. Interesting contrasts emerge, such as the predominance of verbal persuasion verbs in the cause predicate slot of the American English data as opposed to the predominance of physical force verbs in the cause predicate slot of the British English data. We discuss how these and other results create a picture of subtle, yet systematic, differences in meaning construction, and we offer an explanation of these differences as reflecting differently entrenched semantic frames. Keywords: British/American English, Construction Grammar, syntax–lexis interface, collostructional analysis, corpus linguistics, culture, dialectal variation, into-causative construction, meaning potential, semantic frame

1. Introduction Over the past decade, the growing availability of variety-specific corpora has generated an increasing number of studies of regional varieties. With respect to the two major national varieties of English, British English and American, these studies already cover an impressive range of topics as diverse as vocabulary use (Leech and Fallon 1992), differences in conversational style (Tottie 1991), or the distinctive use of modal verbs (Nakamura 1993; Oakes 1992). Moreover, several studies have also documented grammatical differences between the two varieties (compare, among others, Biber 1987; Tottie 2002; Crystal 2004). However, the general

266 Stefanie Wulff, Anatol Stefanowitsch, and Stefan Th. Gries

view still is that, as opposed to the obvious and well-documented phonological and lexical differences, “grammatical differences may be more subtle than the lexical ones” (Tottie 2002: 146), with “very few being general points of syntactic construction” (Crystal 2004: 311), so ultimately “non-trivial syntactic differences between British and American English have been notably harder to find in corpus-based studies” (Kennedy 1998: 193) than phonological, lexical, or discursive differences. In the present chapter, we intend to show how new insights can be gained by adopting a construction-based perspective. Dialectal variation has hardly, if at all, been investigated in Construction Grammar, so we hope to show that this framework can provide new opportunities for this field of research, since it allows us to focus primarily on the syntax–lexis interface rather than phonology, morpho-syntax or lexis. Specifically, our approach, which is based on distinctive collexeme analysis (Gries and Stefanowitsch 2004a), is geared towards pinpointing subtle, yet systematic differences in the way that grammatical patterns found in both varieties are lexically filled. Taking the example of the so-called into-causative construction, we show that, although two such highly similar varieties as British and American English will, of course, share the same inventory of grammatical patterns to a large extent, they may nevertheless differ substantially in meaning construction, i.e., the uses to which they put these patterns.

2. A construction-based approach The theoretical approach we adopt here is Construction Grammar as developed by, among others, Goldberg (1995, 1996) and Lakoff (1987). The central notion of Construction Grammar is the construction, which is defined as follows: A construction is […] a pairing of form and meaning/use such that some aspect of the form or some aspect of the meaning is not strictly predictable from the component parts or from other constructions already established to exist in the language. (Goldberg 1996: 68)

Note that this definition does not distinguish between different kinds of ‘pairings of form and meaning.’ Like some other current theories of language, Construction Grammar no longer assumes a strict division between grammar and lexicon: the linguistic system is viewed as a continuum which covers the whole range from fully lexically specified constructions (e.g., morphemes or words) to partially specified constructions (such as the What’s X doing Y-construction; cf. Kay and Fillmore 1999) to abstract constructions (for example, argument structure constructions or tense and aspect). In order to capture the relationship between constructions of different degrees of abstraction, Construction Grammar makes two fundamental assumptions. First,

Brutal Brits and persuasive Americans 267

since morphemes, words, and the larger, more abstract constructions they occur in are attributed an independent and meaningful status, the words that may occur in a construction must be semantically compatible with the meaning of the construction (or, more precisely, the meaning of the constructional slot into which the word is inserted; cf. Goldberg 1995: 59–66.). Secondly, the so-called Principle of No Synonymy states that “if two constructions are syntactically distinct, they must be semantically or pragmatically distinct” (cf. Goldberg 1995: 67). One major strategy to detect these semantic or pragmatic differences is to investigate those elements of a construction that are variable, i.e. the lexical elements that occupy the ‘free’ slots in a construction. Evidence in favor of the necessity of semantic compatibility and the Principle of No Synonymy originally came from the investigation of classic ‘alternation’ phenomena, such as dative movement, the load/spray alternation, the active/passive alternation, particle movement, and others. Each of these cases has been shown to consist of two distinct constructions, each with its own semantics and/or pragmatics (cf. Goldberg 1995, 2002; Gries and Stefanowitsch 2004a). Moreover, even constructions which were formerly attributed the status of mere surface structure variants can be distinguished on the basis of the words that they typically attract (cf., e.g., Wulff 2006 on go–V vs. go-and-V in English). In the present analysis, we give this approach a slightly new twist: we do not ask to what extent two highly synonymous or syntactically indifferent constructions can be distinguished by looking at their lexical inserts, but rather to what extent we can distinguish British English and American English on the basis of the verbal predicates that British and American speakers insert when they make use of a construction that is part of the inventory of both varieties. In a more technical parlance, we show that the predicates that are attracted to (or repelled by) the construction in question are variety-dependent. While the present study can only begin to outline the potential of a collostructional approach to meaning construction in different cultures, it appears reasonable to start out with a construction that denotes a concept that is fundamental to all cultures. One such concept that lies at the heart of any culture and accordingly continues to inspire (particularly cognitively oriented) linguistic analyses is causation. Therefore, the particular construction that we focus on in the present study to illustrate our line of reasoning is the so-called into-causative construction.

3. Previous studies of the into-causative The into-causative is a construction which is used to denote a causative event by explicitly stating the causer and causee of this event as well as both the action that the

268 Stefanie Wulff, Anatol Stefanowitsch, and Stefan Th. Gries

causer performed to cause the event (the cause predicate, e.g. trick in (1a)) and the resulting action that the causee performs in response to the causer’s action (the result predicate, e.g. employ in (1a)). It is formally restricted in that the result predicate obligatorily follows the preposition into and only occurs in its gerundial form.

(1) a. He tricked me into employing him. b. They were forced into formulating an opinion. c. We conned a grown-up into buying the tickets.

In her analysis of the into-causative, Wierzbicka (1998) notes that the cause predicates occurring in it typically imply, or are at least compatible with, the idea of manipulation. She also points out that the number of potential cause predicates is limited, claiming that while one can trick, manoeuvre, and also talk a person into doing something, one cannot encourage or induce a person into doing something. Similarly, Hunston and Francis’s (2000) corpus-based analysis of the into-causative leads them to conclude that the construction (“pattern”, in their terminology) is associated with “some kind of forcefulness or even coercion” (Hunston and Francis 2000: 16). This is also reflected in the fact that the cause predicates often denote negative emotions (their examples include verbs such as frighten, intimidate, panic, scare, terrify, shock, and shame), or ways of speaking cleverly or deviously (talk, coax, cajole, charm, and browbeat). Gries and Stefanowitsch (2004b) elaborated on earlier studies in several respects: they opted for a maximally objective data selection by including all attestations of the into-causative in their corpus, and they considered both the cause and the result predicates. Most importantly, they provided a quantitative assessment of the association strength between the construction and the verbs that occur in it, as well as of possible interaction between the cause and the result predicates, by employing a technique from the family of methods called collostructional analysis. Since we also employ a technique from this family of methods, we will explain collostructional analysis and the results in more detail in what follows. Collostructional analysis is a family of three methods for determining the direction and the strength of the association between a given construction and the words that occur in one or more of its slots (hence its name, a blend of collocation and construction) (cf. Stefanowitsch and Gries 2003; Gries and Stefanowitsch 2004a, b; Stefanowitsch and Gries 2005). Gries and Stefanowitsch (2004b) apply one of these methods, co-varying collexeme analysis, to identify those pairs of cause and result predicates that are strongly attracted to each other in the into-causative in British English (on the basis of the 1990–2000 volumes of The Guardian). For a brief illustration of this method, consider the association between the verbs bounce and accept in the cause and the result slots of the into-causative, respectively. In order to determine whether these are strongly attracted to each other in the into-causative,

Brutal Brits and persuasive Americans 269

Table 1. 2-by-2 matrix for bounce into accepting in the 1990–2000 volumes of The Guardian bounce other verbs Column totals

accepting

other verbs

Row totals

29 (5.2) 267 (290.8) 296

82 (105.8) 5,910 (5886.2) 5,992

111 6,177 6,288

the following frequencies are required and entered into a two-by-two table as in Table 1 (the italicized values are determined through corpus analysis, the remaining ones can be determined via subtraction): • • • •

the frequency of bounce in the cause slot of the into-causative (111); the frequency of accepting in the result slot of the into-causative (296); the frequency of bounce into accepting in the into-causative (29); the frequency of the into-causative (6,288).

In a first step, the frequencies expected on the basis of chance are computed; they are given in parentheses in Table 1. It is immediately obvious that bounce into accepting occurs much more frequently than expected by chance: it is attracted to the into-causative. To test this result for significance, the table is then subjected to the Fisher-Yates exact test (cf. Stefanowitsch and Gries 2003 for justification), which outputs a p-value providing the probability to obtain the observed table or all more extreme ones. In our example, the p-value is 1.095273E–14. This value means that the probability of getting the result in Table 1 is statistically highly significant (p

Aspects of Meaning Construction (Z 136)

Mental Spaces: Aspects of Meaning Construction in Natural Language

Mental Spaces: Aspects of Meaning Construction in Natural Language

Mental Spaces: Aspects of Meaning Construction in Natural Language

Environmental Aspects of Construction with Waste Materials

The meaning of 'meaning'

The Meaning Of Meaning

Art of Childhood and Adolescence: The Construction of Meaning

The Meaning of Meaning

Художественная галерея № 136. Пармиджанино

Meaning

Meaning

Meaning

Semantic Leaps: Frame-Shifting and Conceptual Blending in Meaning Construction

How Words Mean: Lexical Concepts, Cognitive Models, and Meaning Construction

Meaning

Meaning

Meaning

Meaning

How Words Mean: Lexical Concepts, Cognitive Models, and Meaning Construction

Z

Z

Z

Supervision of Concrete Construction

The Shadow 136 - The Keeper's Gold

The Modes of Meaning

The meaning of relativity

The meaning of relativity

The Meaning Of Language

The Meaning of Masonry

The Meaning of Liff

Aspects of Meaning Construction (Z 136)

Mental Spaces: Aspects of Meaning Construction in Natural Language

Mental Spaces: Aspects of Meaning Construction in Natural Language

Mental Spaces: Aspects of Meaning Construction in Natural Language

Environmental Aspects of Construction with Waste Materials

The meaning of 'meaning'

The Meaning Of Meaning

Art of Childhood and Adolescence: The Construction of Meaning

The Meaning of Meaning

Художественная галерея № 136. Пармиджанино

Meaning

Meaning

Meaning

Semantic Leaps: Frame-Shifting and Conceptual Blending in Meaning Construction

How Words Mean: Lexical Concepts, Cognitive Models, and Meaning Construction

Meaning

Meaning

Meaning

Meaning

How Words Mean: Lexical Concepts, Cognitive Models, and Meaning Construction

Z

Z

Z

Supervision of Concrete Construction

The Shadow 136 - The Keeper's Gold

The Modes of Meaning

The meaning of relativity

The meaning of relativity

The Meaning Of Language

The Meaning of Masonry

The Meaning of Liff

Recommend Documents